ResNet training (+ transfer learning feature).ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test with ResNet in pytorch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from trainer import Trainer\n",
    "from resnet import fineTuneResNet50,lessFilterResNet50, resnet50\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "%matplotlib notebook"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1604, 1003)\n"
     ]
    }
   ],
   "source": [
    "train_df = pd.read_json('Data/combined_training.json')\n",
    "print(train_df.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(75, 75, 3)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "E:\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:1: DeprecationWarning: \n",
      ".ix is deprecated. Please use\n",
      ".loc for label based indexing or\n",
      ".iloc for positional indexing\n",
      "\n",
      "See the documentation here:\n",
      "http://pandas.pydata.org/pandas-docs/stable/indexing.html#ix-indexer-is-deprecated\n",
      "  \"\"\"Entry point for launching an IPython kernel.\n"
     ]
    }
   ],
   "source": [
    "print(np.array(train_df.ix[0, 'band_mixed']).shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[55, 60, 48],\n",
       "        [56, 57, 48],\n",
       "        [53, 54, 48],\n",
       "        ..., \n",
       "        [57, 57, 48],\n",
       "        [56, 56, 47],\n",
       "        [49, 59, 48]],\n",
       "\n",
       "       [[53, 54, 47],\n",
       "        [54, 51, 48],\n",
       "        [55, 54, 47],\n",
       "        ..., \n",
       "        [56, 50, 47],\n",
       "        [52, 58, 47],\n",
       "        [48, 61, 47]],\n",
       "\n",
       "       [[53, 50, 47],\n",
       "        [55, 45, 48],\n",
       "        [58, 54, 48],\n",
       "        ..., \n",
       "        [55, 54, 47],\n",
       "        [54, 53, 47],\n",
       "        [45, 56, 47]],\n",
       "\n",
       "       ..., \n",
       "       [[52, 59, 47],\n",
       "        [53, 59, 47],\n",
       "        [50, 56, 47],\n",
       "        ..., \n",
       "        [52, 63, 47],\n",
       "        [56, 60, 47],\n",
       "        [56, 59, 47]],\n",
       "\n",
       "       [[48, 57, 47],\n",
       "        [54, 57, 47],\n",
       "        [53, 57, 47],\n",
       "        ..., \n",
       "        [55, 59, 47],\n",
       "        [55, 60, 47],\n",
       "        [54, 52, 47]],\n",
       "\n",
       "       [[52, 56, 47],\n",
       "        [55, 58, 47],\n",
       "        [55, 59, 47],\n",
       "        ..., \n",
       "        [55, 57, 47],\n",
       "        [54, 59, 47],\n",
       "        [54, 58, 47]]])"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.array(train_df.ix[0, 'band_mixed'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#model, unused_dict = modResNet18(pretrained=True, skip_headers=['bn2', 'fc'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#print(unused_dict.keys())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import KFold\n",
    "fold_num = 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import torch"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Testing Different Augementation Strategies on Networks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "is_transfer_learning = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1283, 1003) (321, 1003)\n"
     ]
    }
   ],
   "source": [
    "train, valid = train_test_split(train_df, test_size=0.2, random_state=519)\n",
    "train = train.reset_index(drop=True)\n",
    "valid = valid.reset_index(drop=True)\n",
    "print(train.shape, valid.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "E:\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:1: DeprecationWarning: \n",
      ".ix is deprecated. Please use\n",
      ".loc for label based indexing or\n",
      ".iloc for positional indexing\n",
      "\n",
      "See the documentation here:\n",
      "http://pandas.pydata.org/pandas-docs/stable/indexing.html#ix-indexer-is-deprecated\n",
      "  \"\"\"Entry point for launching an IPython kernel.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[[71, 66, 47],\n",
       "        [70, 66, 48],\n",
       "        [67, 66, 48],\n",
       "        ..., \n",
       "        [66, 67, 48],\n",
       "        [65, 67, 47],\n",
       "        [67, 69, 48]],\n",
       "\n",
       "       [[72, 67, 47],\n",
       "        [71, 67, 47],\n",
       "        [65, 65, 48],\n",
       "        ..., \n",
       "        [73, 66, 47],\n",
       "        [68, 69, 47],\n",
       "        [67, 69, 47]],\n",
       "\n",
       "       [[70, 64, 47],\n",
       "        [72, 68, 47],\n",
       "        [70, 66, 48],\n",
       "        ..., \n",
       "        [76, 66, 47],\n",
       "        [72, 66, 47],\n",
       "        [67, 69, 47]],\n",
       "\n",
       "       ..., \n",
       "       [[67, 65, 47],\n",
       "        [64, 63, 47],\n",
       "        [66, 62, 47],\n",
       "        ..., \n",
       "        [69, 62, 47],\n",
       "        [69, 63, 47],\n",
       "        [72, 65, 47]],\n",
       "\n",
       "       [[73, 63, 47],\n",
       "        [70, 59, 47],\n",
       "        [67, 56, 47],\n",
       "        ..., \n",
       "        [60, 65, 47],\n",
       "        [70, 64, 47],\n",
       "        [75, 64, 47]],\n",
       "\n",
       "       [[72, 64, 47],\n",
       "        [73, 65, 47],\n",
       "        [71, 63, 47],\n",
       "        ..., \n",
       "        [68, 67, 47],\n",
       "        [72, 67, 47],\n",
       "        [74, 63, 47]]])"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.array(train.ix[0, 'band_mixed'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1001, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (3049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x00000279B2FD6BE0>, <torchvision.transforms.transforms.Lambda object at 0x00000279B2FD6E48>, <torchvision.transforms.transforms.ToTensor object at 0x00000279B2FD6DA0>]\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x00000279B2FD6C18>, <torchvision.transforms.transforms.ToTensor object at 0x00000279B2FD6C88>]\n"
     ]
    },
    {
     "data": {
      "application/javascript": [
       "/* Put everything inside the global mpl namespace */\n",
       "window.mpl = {};\n",
       "\n",
       "\n",
       "mpl.get_websocket_type = function() {\n",
       "    if (typeof(WebSocket) !== 'undefined') {\n",
       "        return WebSocket;\n",
       "    } else if (typeof(MozWebSocket) !== 'undefined') {\n",
       "        return MozWebSocket;\n",
       "    } else {\n",
       "        alert('Your browser does not have WebSocket support.' +\n",
       "              'Please try Chrome, Safari or Firefox ≥ 6. ' +\n",
       "              'Firefox 4 and 5 are also supported but you ' +\n",
       "              'have to enable WebSockets in about:config.');\n",
       "    };\n",
       "}\n",
       "\n",
       "mpl.figure = function(figure_id, websocket, ondownload, parent_element) {\n",
       "    this.id = figure_id;\n",
       "\n",
       "    this.ws = websocket;\n",
       "\n",
       "    this.supports_binary = (this.ws.binaryType != undefined);\n",
       "\n",
       "    if (!this.supports_binary) {\n",
       "        var warnings = document.getElementById(\"mpl-warnings\");\n",
       "        if (warnings) {\n",
       "            warnings.style.display = 'block';\n",
       "            warnings.textContent = (\n",
       "                \"This browser does not support binary websocket messages. \" +\n",
       "                    \"Performance may be slow.\");\n",
       "        }\n",
       "    }\n",
       "\n",
       "    this.imageObj = new Image();\n",
       "\n",
       "    this.context = undefined;\n",
       "    this.message = undefined;\n",
       "    this.canvas = undefined;\n",
       "    this.rubberband_canvas = undefined;\n",
       "    this.rubberband_context = undefined;\n",
       "    this.format_dropdown = undefined;\n",
       "\n",
       "    this.image_mode = 'full';\n",
       "\n",
       "    this.root = $('<div/>');\n",
       "    this._root_extra_style(this.root)\n",
       "    this.root.attr('style', 'display: inline-block');\n",
       "\n",
       "    $(parent_element).append(this.root);\n",
       "\n",
       "    this._init_header(this);\n",
       "    this._init_canvas(this);\n",
       "    this._init_toolbar(this);\n",
       "\n",
       "    var fig = this;\n",
       "\n",
       "    this.waiting = false;\n",
       "\n",
       "    this.ws.onopen =  function () {\n",
       "            fig.send_message(\"supports_binary\", {value: fig.supports_binary});\n",
       "            fig.send_message(\"send_image_mode\", {});\n",
       "            if (mpl.ratio != 1) {\n",
       "                fig.send_message(\"set_dpi_ratio\", {'dpi_ratio': mpl.ratio});\n",
       "            }\n",
       "            fig.send_message(\"refresh\", {});\n",
       "        }\n",
       "\n",
       "    this.imageObj.onload = function() {\n",
       "            if (fig.image_mode == 'full') {\n",
       "                // Full images could contain transparency (where diff images\n",
       "                // almost always do), so we need to clear the canvas so that\n",
       "                // there is no ghosting.\n",
       "                fig.context.clearRect(0, 0, fig.canvas.width, fig.canvas.height);\n",
       "            }\n",
       "            fig.context.drawImage(fig.imageObj, 0, 0);\n",
       "        };\n",
       "\n",
       "    this.imageObj.onunload = function() {\n",
       "        this.ws.close();\n",
       "    }\n",
       "\n",
       "    this.ws.onmessage = this._make_on_message_function(this);\n",
       "\n",
       "    this.ondownload = ondownload;\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._init_header = function() {\n",
       "    var titlebar = $(\n",
       "        '<div class=\"ui-dialog-titlebar ui-widget-header ui-corner-all ' +\n",
       "        'ui-helper-clearfix\"/>');\n",
       "    var titletext = $(\n",
       "        '<div class=\"ui-dialog-title\" style=\"width: 100%; ' +\n",
       "        'text-align: center; padding: 3px;\"/>');\n",
       "    titlebar.append(titletext)\n",
       "    this.root.append(titlebar);\n",
       "    this.header = titletext[0];\n",
       "}\n",
       "\n",
       "\n",
       "\n",
       "mpl.figure.prototype._canvas_extra_style = function(canvas_div) {\n",
       "\n",
       "}\n",
       "\n",
       "\n",
       "mpl.figure.prototype._root_extra_style = function(canvas_div) {\n",
       "\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._init_canvas = function() {\n",
       "    var fig = this;\n",
       "\n",
       "    var canvas_div = $('<div/>');\n",
       "\n",
       "    canvas_div.attr('style', 'position: relative; clear: both; outline: 0');\n",
       "\n",
       "    function canvas_keyboard_event(event) {\n",
       "        return fig.key_event(event, event['data']);\n",
       "    }\n",
       "\n",
       "    canvas_div.keydown('key_press', canvas_keyboard_event);\n",
       "    canvas_div.keyup('key_release', canvas_keyboard_event);\n",
       "    this.canvas_div = canvas_div\n",
       "    this._canvas_extra_style(canvas_div)\n",
       "    this.root.append(canvas_div);\n",
       "\n",
       "    var canvas = $('<canvas/>');\n",
       "    canvas.addClass('mpl-canvas');\n",
       "    canvas.attr('style', \"left: 0; top: 0; z-index: 0; outline: 0\")\n",
       "\n",
       "    this.canvas = canvas[0];\n",
       "    this.context = canvas[0].getContext(\"2d\");\n",
       "\n",
       "    var backingStore = this.context.backingStorePixelRatio ||\n",
       "\tthis.context.webkitBackingStorePixelRatio ||\n",
       "\tthis.context.mozBackingStorePixelRatio ||\n",
       "\tthis.context.msBackingStorePixelRatio ||\n",
       "\tthis.context.oBackingStorePixelRatio ||\n",
       "\tthis.context.backingStorePixelRatio || 1;\n",
       "\n",
       "    mpl.ratio = (window.devicePixelRatio || 1) / backingStore;\n",
       "\n",
       "    var rubberband = $('<canvas/>');\n",
       "    rubberband.attr('style', \"position: absolute; left: 0; top: 0; z-index: 1;\")\n",
       "\n",
       "    var pass_mouse_events = true;\n",
       "\n",
       "    canvas_div.resizable({\n",
       "        start: function(event, ui) {\n",
       "            pass_mouse_events = false;\n",
       "        },\n",
       "        resize: function(event, ui) {\n",
       "            fig.request_resize(ui.size.width, ui.size.height);\n",
       "        },\n",
       "        stop: function(event, ui) {\n",
       "            pass_mouse_events = true;\n",
       "            fig.request_resize(ui.size.width, ui.size.height);\n",
       "        },\n",
       "    });\n",
       "\n",
       "    function mouse_event_fn(event) {\n",
       "        if (pass_mouse_events)\n",
       "            return fig.mouse_event(event, event['data']);\n",
       "    }\n",
       "\n",
       "    rubberband.mousedown('button_press', mouse_event_fn);\n",
       "    rubberband.mouseup('button_release', mouse_event_fn);\n",
       "    // Throttle sequential mouse events to 1 every 20ms.\n",
       "    rubberband.mousemove('motion_notify', mouse_event_fn);\n",
       "\n",
       "    rubberband.mouseenter('figure_enter', mouse_event_fn);\n",
       "    rubberband.mouseleave('figure_leave', mouse_event_fn);\n",
       "\n",
       "    canvas_div.on(\"wheel\", function (event) {\n",
       "        event = event.originalEvent;\n",
       "        event['data'] = 'scroll'\n",
       "        if (event.deltaY < 0) {\n",
       "            event.step = 1;\n",
       "        } else {\n",
       "            event.step = -1;\n",
       "        }\n",
       "        mouse_event_fn(event);\n",
       "    });\n",
       "\n",
       "    canvas_div.append(canvas);\n",
       "    canvas_div.append(rubberband);\n",
       "\n",
       "    this.rubberband = rubberband;\n",
       "    this.rubberband_canvas = rubberband[0];\n",
       "    this.rubberband_context = rubberband[0].getContext(\"2d\");\n",
       "    this.rubberband_context.strokeStyle = \"#000000\";\n",
       "\n",
       "    this._resize_canvas = function(width, height) {\n",
       "        // Keep the size of the canvas, canvas container, and rubber band\n",
       "        // canvas in synch.\n",
       "        canvas_div.css('width', width)\n",
       "        canvas_div.css('height', height)\n",
       "\n",
       "        canvas.attr('width', width * mpl.ratio);\n",
       "        canvas.attr('height', height * mpl.ratio);\n",
       "        canvas.attr('style', 'width: ' + width + 'px; height: ' + height + 'px;');\n",
       "\n",
       "        rubberband.attr('width', width);\n",
       "        rubberband.attr('height', height);\n",
       "    }\n",
       "\n",
       "    // Set the figure to an initial 600x600px, this will subsequently be updated\n",
       "    // upon first draw.\n",
       "    this._resize_canvas(600, 600);\n",
       "\n",
       "    // Disable right mouse context menu.\n",
       "    $(this.rubberband_canvas).bind(\"contextmenu\",function(e){\n",
       "        return false;\n",
       "    });\n",
       "\n",
       "    function set_focus () {\n",
       "        canvas.focus();\n",
       "        canvas_div.focus();\n",
       "    }\n",
       "\n",
       "    window.setTimeout(set_focus, 100);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._init_toolbar = function() {\n",
       "    var fig = this;\n",
       "\n",
       "    var nav_element = $('<div/>')\n",
       "    nav_element.attr('style', 'width: 100%');\n",
       "    this.root.append(nav_element);\n",
       "\n",
       "    // Define a callback function for later on.\n",
       "    function toolbar_event(event) {\n",
       "        return fig.toolbar_button_onclick(event['data']);\n",
       "    }\n",
       "    function toolbar_mouse_event(event) {\n",
       "        return fig.toolbar_button_onmouseover(event['data']);\n",
       "    }\n",
       "\n",
       "    for(var toolbar_ind in mpl.toolbar_items) {\n",
       "        var name = mpl.toolbar_items[toolbar_ind][0];\n",
       "        var tooltip = mpl.toolbar_items[toolbar_ind][1];\n",
       "        var image = mpl.toolbar_items[toolbar_ind][2];\n",
       "        var method_name = mpl.toolbar_items[toolbar_ind][3];\n",
       "\n",
       "        if (!name) {\n",
       "            // put a spacer in here.\n",
       "            continue;\n",
       "        }\n",
       "        var button = $('<button/>');\n",
       "        button.addClass('ui-button ui-widget ui-state-default ui-corner-all ' +\n",
       "                        'ui-button-icon-only');\n",
       "        button.attr('role', 'button');\n",
       "        button.attr('aria-disabled', 'false');\n",
       "        button.click(method_name, toolbar_event);\n",
       "        button.mouseover(tooltip, toolbar_mouse_event);\n",
       "\n",
       "        var icon_img = $('<span/>');\n",
       "        icon_img.addClass('ui-button-icon-primary ui-icon');\n",
       "        icon_img.addClass(image);\n",
       "        icon_img.addClass('ui-corner-all');\n",
       "\n",
       "        var tooltip_span = $('<span/>');\n",
       "        tooltip_span.addClass('ui-button-text');\n",
       "        tooltip_span.html(tooltip);\n",
       "\n",
       "        button.append(icon_img);\n",
       "        button.append(tooltip_span);\n",
       "\n",
       "        nav_element.append(button);\n",
       "    }\n",
       "\n",
       "    var fmt_picker_span = $('<span/>');\n",
       "\n",
       "    var fmt_picker = $('<select/>');\n",
       "    fmt_picker.addClass('mpl-toolbar-option ui-widget ui-widget-content');\n",
       "    fmt_picker_span.append(fmt_picker);\n",
       "    nav_element.append(fmt_picker_span);\n",
       "    this.format_dropdown = fmt_picker[0];\n",
       "\n",
       "    for (var ind in mpl.extensions) {\n",
       "        var fmt = mpl.extensions[ind];\n",
       "        var option = $(\n",
       "            '<option/>', {selected: fmt === mpl.default_extension}).html(fmt);\n",
       "        fmt_picker.append(option)\n",
       "    }\n",
       "\n",
       "    // Add hover states to the ui-buttons\n",
       "    $( \".ui-button\" ).hover(\n",
       "        function() { $(this).addClass(\"ui-state-hover\");},\n",
       "        function() { $(this).removeClass(\"ui-state-hover\");}\n",
       "    );\n",
       "\n",
       "    var status_bar = $('<span class=\"mpl-message\"/>');\n",
       "    nav_element.append(status_bar);\n",
       "    this.message = status_bar[0];\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.request_resize = function(x_pixels, y_pixels) {\n",
       "    // Request matplotlib to resize the figure. Matplotlib will then trigger a resize in the client,\n",
       "    // which will in turn request a refresh of the image.\n",
       "    this.send_message('resize', {'width': x_pixels, 'height': y_pixels});\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.send_message = function(type, properties) {\n",
       "    properties['type'] = type;\n",
       "    properties['figure_id'] = this.id;\n",
       "    this.ws.send(JSON.stringify(properties));\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.send_draw_message = function() {\n",
       "    if (!this.waiting) {\n",
       "        this.waiting = true;\n",
       "        this.ws.send(JSON.stringify({type: \"draw\", figure_id: this.id}));\n",
       "    }\n",
       "}\n",
       "\n",
       "\n",
       "mpl.figure.prototype.handle_save = function(fig, msg) {\n",
       "    var format_dropdown = fig.format_dropdown;\n",
       "    var format = format_dropdown.options[format_dropdown.selectedIndex].value;\n",
       "    fig.ondownload(fig, format);\n",
       "}\n",
       "\n",
       "\n",
       "mpl.figure.prototype.handle_resize = function(fig, msg) {\n",
       "    var size = msg['size'];\n",
       "    if (size[0] != fig.canvas.width || size[1] != fig.canvas.height) {\n",
       "        fig._resize_canvas(size[0], size[1]);\n",
       "        fig.send_message(\"refresh\", {});\n",
       "    };\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_rubberband = function(fig, msg) {\n",
       "    var x0 = msg['x0'] / mpl.ratio;\n",
       "    var y0 = (fig.canvas.height - msg['y0']) / mpl.ratio;\n",
       "    var x1 = msg['x1'] / mpl.ratio;\n",
       "    var y1 = (fig.canvas.height - msg['y1']) / mpl.ratio;\n",
       "    x0 = Math.floor(x0) + 0.5;\n",
       "    y0 = Math.floor(y0) + 0.5;\n",
       "    x1 = Math.floor(x1) + 0.5;\n",
       "    y1 = Math.floor(y1) + 0.5;\n",
       "    var min_x = Math.min(x0, x1);\n",
       "    var min_y = Math.min(y0, y1);\n",
       "    var width = Math.abs(x1 - x0);\n",
       "    var height = Math.abs(y1 - y0);\n",
       "\n",
       "    fig.rubberband_context.clearRect(\n",
       "        0, 0, fig.canvas.width, fig.canvas.height);\n",
       "\n",
       "    fig.rubberband_context.strokeRect(min_x, min_y, width, height);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_figure_label = function(fig, msg) {\n",
       "    // Updates the figure title.\n",
       "    fig.header.textContent = msg['label'];\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_cursor = function(fig, msg) {\n",
       "    var cursor = msg['cursor'];\n",
       "    switch(cursor)\n",
       "    {\n",
       "    case 0:\n",
       "        cursor = 'pointer';\n",
       "        break;\n",
       "    case 1:\n",
       "        cursor = 'default';\n",
       "        break;\n",
       "    case 2:\n",
       "        cursor = 'crosshair';\n",
       "        break;\n",
       "    case 3:\n",
       "        cursor = 'move';\n",
       "        break;\n",
       "    }\n",
       "    fig.rubberband_canvas.style.cursor = cursor;\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_message = function(fig, msg) {\n",
       "    fig.message.textContent = msg['message'];\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_draw = function(fig, msg) {\n",
       "    // Request the server to send over a new figure.\n",
       "    fig.send_draw_message();\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_image_mode = function(fig, msg) {\n",
       "    fig.image_mode = msg['mode'];\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.updated_canvas_event = function() {\n",
       "    // Called whenever the canvas gets updated.\n",
       "    this.send_message(\"ack\", {});\n",
       "}\n",
       "\n",
       "// A function to construct a web socket function for onmessage handling.\n",
       "// Called in the figure constructor.\n",
       "mpl.figure.prototype._make_on_message_function = function(fig) {\n",
       "    return function socket_on_message(evt) {\n",
       "        if (evt.data instanceof Blob) {\n",
       "            /* FIXME: We get \"Resource interpreted as Image but\n",
       "             * transferred with MIME type text/plain:\" errors on\n",
       "             * Chrome.  But how to set the MIME type?  It doesn't seem\n",
       "             * to be part of the websocket stream */\n",
       "            evt.data.type = \"image/png\";\n",
       "\n",
       "            /* Free the memory for the previous frames */\n",
       "            if (fig.imageObj.src) {\n",
       "                (window.URL || window.webkitURL).revokeObjectURL(\n",
       "                    fig.imageObj.src);\n",
       "            }\n",
       "\n",
       "            fig.imageObj.src = (window.URL || window.webkitURL).createObjectURL(\n",
       "                evt.data);\n",
       "            fig.updated_canvas_event();\n",
       "            fig.waiting = false;\n",
       "            return;\n",
       "        }\n",
       "        else if (typeof evt.data === 'string' && evt.data.slice(0, 21) == \"data:image/png;base64\") {\n",
       "            fig.imageObj.src = evt.data;\n",
       "            fig.updated_canvas_event();\n",
       "            fig.waiting = false;\n",
       "            return;\n",
       "        }\n",
       "\n",
       "        var msg = JSON.parse(evt.data);\n",
       "        var msg_type = msg['type'];\n",
       "\n",
       "        // Call the  \"handle_{type}\" callback, which takes\n",
       "        // the figure and JSON message as its only arguments.\n",
       "        try {\n",
       "            var callback = fig[\"handle_\" + msg_type];\n",
       "        } catch (e) {\n",
       "            console.log(\"No handler for the '\" + msg_type + \"' message type: \", msg);\n",
       "            return;\n",
       "        }\n",
       "\n",
       "        if (callback) {\n",
       "            try {\n",
       "                // console.log(\"Handling '\" + msg_type + \"' message: \", msg);\n",
       "                callback(fig, msg);\n",
       "            } catch (e) {\n",
       "                console.log(\"Exception inside the 'handler_\" + msg_type + \"' callback:\", e, e.stack, msg);\n",
       "            }\n",
       "        }\n",
       "    };\n",
       "}\n",
       "\n",
       "// from http://stackoverflow.com/questions/1114465/getting-mouse-location-in-canvas\n",
       "mpl.findpos = function(e) {\n",
       "    //this section is from http://www.quirksmode.org/js/events_properties.html\n",
       "    var targ;\n",
       "    if (!e)\n",
       "        e = window.event;\n",
       "    if (e.target)\n",
       "        targ = e.target;\n",
       "    else if (e.srcElement)\n",
       "        targ = e.srcElement;\n",
       "    if (targ.nodeType == 3) // defeat Safari bug\n",
       "        targ = targ.parentNode;\n",
       "\n",
       "    // jQuery normalizes the pageX and pageY\n",
       "    // pageX,Y are the mouse positions relative to the document\n",
       "    // offset() returns the position of the element relative to the document\n",
       "    var x = e.pageX - $(targ).offset().left;\n",
       "    var y = e.pageY - $(targ).offset().top;\n",
       "\n",
       "    return {\"x\": x, \"y\": y};\n",
       "};\n",
       "\n",
       "/*\n",
       " * return a copy of an object with only non-object keys\n",
       " * we need this to avoid circular references\n",
       " * http://stackoverflow.com/a/24161582/3208463\n",
       " */\n",
       "function simpleKeys (original) {\n",
       "  return Object.keys(original).reduce(function (obj, key) {\n",
       "    if (typeof original[key] !== 'object')\n",
       "        obj[key] = original[key]\n",
       "    return obj;\n",
       "  }, {});\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.mouse_event = function(event, name) {\n",
       "    var canvas_pos = mpl.findpos(event)\n",
       "\n",
       "    if (name === 'button_press')\n",
       "    {\n",
       "        this.canvas.focus();\n",
       "        this.canvas_div.focus();\n",
       "    }\n",
       "\n",
       "    var x = canvas_pos.x * mpl.ratio;\n",
       "    var y = canvas_pos.y * mpl.ratio;\n",
       "\n",
       "    this.send_message(name, {x: x, y: y, button: event.button,\n",
       "                             step: event.step,\n",
       "                             guiEvent: simpleKeys(event)});\n",
       "\n",
       "    /* This prevents the web browser from automatically changing to\n",
       "     * the text insertion cursor when the button is pressed.  We want\n",
       "     * to control all of the cursor setting manually through the\n",
       "     * 'cursor' event from matplotlib */\n",
       "    event.preventDefault();\n",
       "    return false;\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._key_event_extra = function(event, name) {\n",
       "    // Handle any extra behaviour associated with a key event\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.key_event = function(event, name) {\n",
       "\n",
       "    // Prevent repeat events\n",
       "    if (name == 'key_press')\n",
       "    {\n",
       "        if (event.which === this._key)\n",
       "            return;\n",
       "        else\n",
       "            this._key = event.which;\n",
       "    }\n",
       "    if (name == 'key_release')\n",
       "        this._key = null;\n",
       "\n",
       "    var value = '';\n",
       "    if (event.ctrlKey && event.which != 17)\n",
       "        value += \"ctrl+\";\n",
       "    if (event.altKey && event.which != 18)\n",
       "        value += \"alt+\";\n",
       "    if (event.shiftKey && event.which != 16)\n",
       "        value += \"shift+\";\n",
       "\n",
       "    value += 'k';\n",
       "    value += event.which.toString();\n",
       "\n",
       "    this._key_event_extra(event, name);\n",
       "\n",
       "    this.send_message(name, {key: value,\n",
       "                             guiEvent: simpleKeys(event)});\n",
       "    return false;\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.toolbar_button_onclick = function(name) {\n",
       "    if (name == 'download') {\n",
       "        this.handle_save(this, null);\n",
       "    } else {\n",
       "        this.send_message(\"toolbar_button\", {name: name});\n",
       "    }\n",
       "};\n",
       "\n",
       "mpl.figure.prototype.toolbar_button_onmouseover = function(tooltip) {\n",
       "    this.message.textContent = tooltip;\n",
       "};\n",
       "mpl.toolbar_items = [[\"Home\", \"Reset original view\", \"fa fa-home icon-home\", \"home\"], [\"Back\", \"Back to  previous view\", \"fa fa-arrow-left icon-arrow-left\", \"back\"], [\"Forward\", \"Forward to next view\", \"fa fa-arrow-right icon-arrow-right\", \"forward\"], [\"\", \"\", \"\", \"\"], [\"Pan\", \"Pan axes with left mouse, zoom with right\", \"fa fa-arrows icon-move\", \"pan\"], [\"Zoom\", \"Zoom to rectangle\", \"fa fa-square-o icon-check-empty\", \"zoom\"], [\"\", \"\", \"\", \"\"], [\"Download\", \"Download plot\", \"fa fa-floppy-o icon-save\", \"download\"]];\n",
       "\n",
       "mpl.extensions = [\"eps\", \"jpeg\", \"pdf\", \"png\", \"ps\", \"raw\", \"svg\", \"tif\"];\n",
       "\n",
       "mpl.default_extension = \"png\";var comm_websocket_adapter = function(comm) {\n",
       "    // Create a \"websocket\"-like object which calls the given IPython comm\n",
       "    // object with the appropriate methods. Currently this is a non binary\n",
       "    // socket, so there is still some room for performance tuning.\n",
       "    var ws = {};\n",
       "\n",
       "    ws.close = function() {\n",
       "        comm.close()\n",
       "    };\n",
       "    ws.send = function(m) {\n",
       "        //console.log('sending', m);\n",
       "        comm.send(m);\n",
       "    };\n",
       "    // Register the callback with on_msg.\n",
       "    comm.on_msg(function(msg) {\n",
       "        //console.log('receiving', msg['content']['data'], msg);\n",
       "        // Pass the mpl event to the overriden (by mpl) onmessage function.\n",
       "        ws.onmessage(msg['content']['data'])\n",
       "    });\n",
       "    return ws;\n",
       "}\n",
       "\n",
       "mpl.mpl_figure_comm = function(comm, msg) {\n",
       "    // This is the function which gets called when the mpl process\n",
       "    // starts-up an IPython Comm through the \"matplotlib\" channel.\n",
       "\n",
       "    var id = msg.content.data.id;\n",
       "    // Get hold of the div created by the display call when the Comm\n",
       "    // socket was opened in Python.\n",
       "    var element = $(\"#\" + id);\n",
       "    var ws_proxy = comm_websocket_adapter(comm)\n",
       "\n",
       "    function ondownload(figure, format) {\n",
       "        window.open(figure.imageObj.src);\n",
       "    }\n",
       "\n",
       "    var fig = new mpl.figure(id, ws_proxy,\n",
       "                           ondownload,\n",
       "                           element.get(0));\n",
       "\n",
       "    // Call onopen now - mpl needs it, as it is assuming we've passed it a real\n",
       "    // web socket which is closed, not our websocket->open comm proxy.\n",
       "    ws_proxy.onopen();\n",
       "\n",
       "    fig.parent_element = element.get(0);\n",
       "    fig.cell_info = mpl.find_output_cell(\"<div id='\" + id + \"'></div>\");\n",
       "    if (!fig.cell_info) {\n",
       "        console.error(\"Failed to find cell for figure\", id, fig);\n",
       "        return;\n",
       "    }\n",
       "\n",
       "    var output_index = fig.cell_info[2]\n",
       "    var cell = fig.cell_info[0];\n",
       "\n",
       "};\n",
       "\n",
       "mpl.figure.prototype.handle_close = function(fig, msg) {\n",
       "    var width = fig.canvas.width/mpl.ratio\n",
       "    fig.root.unbind('remove')\n",
       "\n",
       "    // Update the output cell to use the data from the current canvas.\n",
       "    fig.push_to_output();\n",
       "    var dataURL = fig.canvas.toDataURL();\n",
       "    // Re-enable the keyboard manager in IPython - without this line, in FF,\n",
       "    // the notebook keyboard shortcuts fail.\n",
       "    IPython.keyboard_manager.enable()\n",
       "    $(fig.parent_element).html('<img src=\"' + dataURL + '\" width=\"' + width + '\">');\n",
       "    fig.close_ws(fig, msg);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.close_ws = function(fig, msg){\n",
       "    fig.send_message('closing', msg);\n",
       "    // fig.ws.close()\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.push_to_output = function(remove_interactive) {\n",
       "    // Turn the data on the canvas into data in the output cell.\n",
       "    var width = this.canvas.width/mpl.ratio\n",
       "    var dataURL = this.canvas.toDataURL();\n",
       "    this.cell_info[1]['text/html'] = '<img src=\"' + dataURL + '\" width=\"' + width + '\">';\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.updated_canvas_event = function() {\n",
       "    // Tell IPython that the notebook contents must change.\n",
       "    IPython.notebook.set_dirty(true);\n",
       "    this.send_message(\"ack\", {});\n",
       "    var fig = this;\n",
       "    // Wait a second, then push the new image to the DOM so\n",
       "    // that it is saved nicely (might be nice to debounce this).\n",
       "    setTimeout(function () { fig.push_to_output() }, 1000);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._init_toolbar = function() {\n",
       "    var fig = this;\n",
       "\n",
       "    var nav_element = $('<div/>')\n",
       "    nav_element.attr('style', 'width: 100%');\n",
       "    this.root.append(nav_element);\n",
       "\n",
       "    // Define a callback function for later on.\n",
       "    function toolbar_event(event) {\n",
       "        return fig.toolbar_button_onclick(event['data']);\n",
       "    }\n",
       "    function toolbar_mouse_event(event) {\n",
       "        return fig.toolbar_button_onmouseover(event['data']);\n",
       "    }\n",
       "\n",
       "    for(var toolbar_ind in mpl.toolbar_items){\n",
       "        var name = mpl.toolbar_items[toolbar_ind][0];\n",
       "        var tooltip = mpl.toolbar_items[toolbar_ind][1];\n",
       "        var image = mpl.toolbar_items[toolbar_ind][2];\n",
       "        var method_name = mpl.toolbar_items[toolbar_ind][3];\n",
       "\n",
       "        if (!name) { continue; };\n",
       "\n",
       "        var button = $('<button class=\"btn btn-default\" href=\"#\" title=\"' + name + '\"><i class=\"fa ' + image + ' fa-lg\"></i></button>');\n",
       "        button.click(method_name, toolbar_event);\n",
       "        button.mouseover(tooltip, toolbar_mouse_event);\n",
       "        nav_element.append(button);\n",
       "    }\n",
       "\n",
       "    // Add the status bar.\n",
       "    var status_bar = $('<span class=\"mpl-message\" style=\"text-align:right; float: right;\"/>');\n",
       "    nav_element.append(status_bar);\n",
       "    this.message = status_bar[0];\n",
       "\n",
       "    // Add the close button to the window.\n",
       "    var buttongrp = $('<div class=\"btn-group inline pull-right\"></div>');\n",
       "    var button = $('<button class=\"btn btn-mini btn-primary\" href=\"#\" title=\"Stop Interaction\"><i class=\"fa fa-power-off icon-remove icon-large\"></i></button>');\n",
       "    button.click(function (evt) { fig.handle_close(fig, {}); } );\n",
       "    button.mouseover('Stop Interaction', toolbar_mouse_event);\n",
       "    buttongrp.append(button);\n",
       "    var titlebar = this.root.find($('.ui-dialog-titlebar'));\n",
       "    titlebar.prepend(buttongrp);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._root_extra_style = function(el){\n",
       "    var fig = this\n",
       "    el.on(\"remove\", function(){\n",
       "\tfig.close_ws(fig, {});\n",
       "    });\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._canvas_extra_style = function(el){\n",
       "    // this is important to make the div 'focusable\n",
       "    el.attr('tabindex', 0)\n",
       "    // reach out to IPython and tell the keyboard manager to turn it's self\n",
       "    // off when our div gets focus\n",
       "\n",
       "    // location in version 3\n",
       "    if (IPython.notebook.keyboard_manager) {\n",
       "        IPython.notebook.keyboard_manager.register_events(el);\n",
       "    }\n",
       "    else {\n",
       "        // location in version 2\n",
       "        IPython.keyboard_manager.register_events(el);\n",
       "    }\n",
       "\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._key_event_extra = function(event, name) {\n",
       "    var manager = IPython.notebook.keyboard_manager;\n",
       "    if (!manager)\n",
       "        manager = IPython.keyboard_manager;\n",
       "\n",
       "    // Check for shift+enter\n",
       "    if (event.shiftKey && event.which == 13) {\n",
       "        this.canvas_div.blur();\n",
       "        // select the cell after this one\n",
       "        var index = IPython.notebook.find_cell_index(this.cell_info[0]);\n",
       "        IPython.notebook.select(index + 1);\n",
       "    }\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_save = function(fig, msg) {\n",
       "    fig.ondownload(fig, null);\n",
       "}\n",
       "\n",
       "\n",
       "mpl.find_output_cell = function(html_output) {\n",
       "    // Return the cell and output element which can be found *uniquely* in the notebook.\n",
       "    // Note - this is a bit hacky, but it is done because the \"notebook_saving.Notebook\"\n",
       "    // IPython event is triggered only after the cells have been serialised, which for\n",
       "    // our purposes (turning an active figure into a static one), is too late.\n",
       "    var cells = IPython.notebook.get_cells();\n",
       "    var ncells = cells.length;\n",
       "    for (var i=0; i<ncells; i++) {\n",
       "        var cell = cells[i];\n",
       "        if (cell.cell_type === 'code'){\n",
       "            for (var j=0; j<cell.output_area.outputs.length; j++) {\n",
       "                var data = cell.output_area.outputs[j];\n",
       "                if (data.data) {\n",
       "                    // IPython >= 3 moved mimebundle to data attribute of output\n",
       "                    data = data.data;\n",
       "                }\n",
       "                if (data['text/html'] == html_output) {\n",
       "                    return [cell, data, j];\n",
       "                }\n",
       "            }\n",
       "        }\n",
       "    }\n",
       "}\n",
       "\n",
       "// Register the function which deals with the matplotlib target/channel.\n",
       "// The kernel may be null if the page has been refreshed.\n",
       "if (IPython.notebook.kernel != null) {\n",
       "    IPython.notebook.kernel.comm_manager.register_target('matplotlib', mpl.mpl_figure_comm);\n",
       "}\n"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<img src=\"data:image/png;base64,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\" width=\"1000\">"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 10: Train set: Average loss: 0.8930, Accuracy: 721/1283 (56.20%)\n",
      "Valid set: Average loss: 0.7813, Accuracy: 162/321 (50.47%)\n",
      "best_loss updated:  0.7813359101613363\n",
      "epoch= 20: Train set: Average loss: 0.7403, Accuracy: 709/1283 (55.26%)\n",
      "Valid set: Average loss: 0.7312, Accuracy: 166/321 (51.71%)\n",
      "best_loss updated:  0.7312094767888387\n",
      "epoch= 30: Train set: Average loss: 0.6437, Accuracy: 822/1283 (64.07%)\n",
      "Valid set: Average loss: 0.6831, Accuracy: 186/321 (57.94%)\n",
      "best_loss updated:  0.6831038196881613\n",
      "epoch= 40: Train set: Average loss: 0.6362, Accuracy: 796/1283 (62.04%)\n",
      "Valid set: Average loss: 0.6749, Accuracy: 183/321 (57.01%)\n",
      "best_loss updated:  0.6748588482538859\n",
      "epoch= 50: Train set: Average loss: 0.6178, Accuracy: 891/1283 (69.45%)\n",
      "Valid set: Average loss: 0.7641, Accuracy: 169/321 (52.65%)\n",
      "epoch= 60: Train set: Average loss: 0.5669, Accuracy: 885/1283 (68.98%)\n",
      "Valid set: Average loss: 0.6543, Accuracy: 195/321 (60.75%)\n",
      "best_loss updated:  0.654273251692454\n",
      "epoch= 70: Train set: Average loss: 0.6295, Accuracy: 804/1283 (62.67%)\n",
      "Valid set: Average loss: 1.4560, Accuracy: 172/321 (53.58%)\n",
      "epoch= 80: Train set: Average loss: 0.6303, Accuracy: 862/1283 (67.19%)\n",
      "Valid set: Average loss: 0.6991, Accuracy: 178/321 (55.45%)\n",
      "epoch= 90: Train set: Average loss: 0.5593, Accuracy: 864/1283 (67.34%)\n",
      "Valid set: Average loss: 0.5573, Accuracy: 218/321 (67.91%)\n",
      "best_loss updated:  0.5572893818219503\n",
      "epoch= 100: Train set: Average loss: 0.5993, Accuracy: 927/1283 (72.25%)\n",
      "Valid set: Average loss: 0.5856, Accuracy: 221/321 (68.85%)\n",
      "epoch= 110: Train set: Average loss: 0.5140, Accuracy: 971/1283 (75.68%)\n",
      "Valid set: Average loss: 0.7224, Accuracy: 176/321 (54.83%)\n",
      "epoch= 120: Train set: Average loss: 0.4414, Accuracy: 1040/1283 (81.06%)\n",
      "Valid set: Average loss: 0.4419, Accuracy: 255/321 (79.44%)\n",
      "best_loss updated:  0.44194472829500836\n",
      "epoch= 130: Train set: Average loss: 0.5333, Accuracy: 1020/1283 (79.50%)\n",
      "Valid set: Average loss: 0.5235, Accuracy: 228/321 (71.03%)\n",
      "epoch= 140: Train set: Average loss: 0.4048, Accuracy: 1032/1283 (80.44%)\n",
      "Valid set: Average loss: 0.6834, Accuracy: 203/321 (63.24%)\n",
      "epoch= 150: Train set: Average loss: 0.5599, Accuracy: 1003/1283 (78.18%)\n",
      "Valid set: Average loss: 0.6223, Accuracy: 188/321 (58.57%)\n",
      "epoch= 160: Train set: Average loss: 0.5212, Accuracy: 991/1283 (77.24%)\n",
      "Valid set: Average loss: 0.5115, Accuracy: 235/321 (73.21%)\n",
      "epoch= 170: Train set: Average loss: 0.3720, Accuracy: 1078/1283 (84.02%)\n",
      "Valid set: Average loss: 0.4051, Accuracy: 267/321 (83.18%)\n",
      "best_loss updated:  0.40505604942639667\n",
      "epoch= 180: Train set: Average loss: 0.4063, Accuracy: 1052/1283 (82.00%)\n",
      "Valid set: Average loss: 0.4720, Accuracy: 250/321 (77.88%)\n",
      "epoch= 190: Train set: Average loss: 0.4645, Accuracy: 1056/1283 (82.31%)\n",
      "Valid set: Average loss: 0.7788, Accuracy: 207/321 (64.49%)\n",
      "epoch= 200: Train set: Average loss: 0.4198, Accuracy: 1031/1283 (80.36%)\n",
      "Valid set: Average loss: 0.5723, Accuracy: 222/321 (69.16%)\n",
      "epoch= 210: Train set: Average loss: 0.4655, Accuracy: 979/1283 (76.31%)\n",
      "Valid set: Average loss: 0.5052, Accuracy: 241/321 (75.08%)\n",
      "epoch= 220: Train set: Average loss: 0.4731, Accuracy: 1012/1283 (78.88%)\n",
      "Valid set: Average loss: 0.5131, Accuracy: 254/321 (79.13%)\n",
      "epoch= 230: Train set: Average loss: 0.4245, Accuracy: 1037/1283 (80.83%)\n",
      "Valid set: Average loss: 0.5241, Accuracy: 247/321 (76.95%)\n",
      "epoch= 240: Train set: Average loss: 0.3538, Accuracy: 1080/1283 (84.18%)\n",
      "Valid set: Average loss: 0.4487, Accuracy: 257/321 (80.06%)\n",
      "epoch= 250: Train set: Average loss: 0.3838, Accuracy: 1072/1283 (83.55%)\n",
      "Valid set: Average loss: 0.5198, Accuracy: 250/321 (77.88%)\n",
      "epoch= 260: Train set: Average loss: 0.4856, Accuracy: 981/1283 (76.46%)\n",
      "Valid set: Average loss: 0.7498, Accuracy: 194/321 (60.44%)\n",
      "epoch= 270: Train set: Average loss: 0.3901, Accuracy: 1046/1283 (81.53%)\n",
      "Valid set: Average loss: 0.4523, Accuracy: 257/321 (80.06%)\n",
      "epoch= 280: Train set: Average loss: 0.4973, Accuracy: 1067/1283 (83.16%)\n",
      "Valid set: Average loss: 0.5565, Accuracy: 235/321 (73.21%)\n",
      "epoch= 290: Train set: Average loss: 0.3630, Accuracy: 1083/1283 (84.41%)\n",
      "Valid set: Average loss: 0.5287, Accuracy: 250/321 (77.88%)\n",
      "epoch= 300: Train set: Average loss: 0.3922, Accuracy: 1040/1283 (81.06%)\n",
      "Valid set: Average loss: 0.5107, Accuracy: 249/321 (77.57%)\n",
      "epoch= 310: Train set: Average loss: 0.4673, Accuracy: 1094/1283 (85.27%)\n",
      "Valid set: Average loss: 0.3882, Accuracy: 274/321 (85.36%)\n",
      "best_loss updated:  0.38816121220588684\n",
      "epoch= 320: Train set: Average loss: 0.5103, Accuracy: 1061/1283 (82.70%)\n",
      "Valid set: Average loss: 0.4692, Accuracy: 257/321 (80.06%)\n",
      "epoch= 330: Train set: Average loss: 0.3948, Accuracy: 1072/1283 (83.55%)\n",
      "Valid set: Average loss: 0.4345, Accuracy: 263/321 (81.93%)\n",
      "epoch= 340: Train set: Average loss: 0.4328, Accuracy: 1102/1283 (85.89%)\n",
      "Valid set: Average loss: 0.3954, Accuracy: 271/321 (84.42%)\n",
      "epoch= 350: Train set: Average loss: 0.4350, Accuracy: 1118/1283 (87.14%)\n",
      "Valid set: Average loss: 0.4022, Accuracy: 265/321 (82.55%)\n",
      "epoch= 360: Train set: Average loss: 0.3402, Accuracy: 1103/1283 (85.97%)\n",
      "Valid set: Average loss: 0.4165, Accuracy: 264/321 (82.24%)\n",
      "epoch= 370: Train set: Average loss: 0.3678, Accuracy: 1090/1283 (84.96%)\n",
      "Valid set: Average loss: 0.4404, Accuracy: 262/321 (81.62%)\n",
      "epoch= 380: Train set: Average loss: 0.4484, Accuracy: 1115/1283 (86.91%)\n",
      "Valid set: Average loss: 0.3980, Accuracy: 274/321 (85.36%)\n",
      "epoch= 390: Train set: Average loss: 0.3849, Accuracy: 1119/1283 (87.22%)\n",
      "Valid set: Average loss: 0.3934, Accuracy: 268/321 (83.49%)\n",
      "epoch= 400: Train set: Average loss: 0.4229, Accuracy: 1080/1283 (84.18%)\n",
      "Valid set: Average loss: 0.4421, Accuracy: 262/321 (81.62%)\n",
      "epoch= 410: Train set: Average loss: 0.2849, Accuracy: 1122/1283 (87.45%)\n",
      "Valid set: Average loss: 0.3654, Accuracy: 277/321 (86.29%)\n",
      "best_loss updated:  0.3654302457968394\n",
      "epoch= 420: Train set: Average loss: 0.2869, Accuracy: 1128/1283 (87.92%)\n",
      "Valid set: Average loss: 0.4002, Accuracy: 268/321 (83.49%)\n",
      "epoch= 430: Train set: Average loss: 0.3200, Accuracy: 1117/1283 (87.06%)\n",
      "Valid set: Average loss: 0.3818, Accuracy: 271/321 (84.42%)\n",
      "epoch= 440: Train set: Average loss: 0.4026, Accuracy: 1125/1283 (87.69%)\n",
      "Valid set: Average loss: 0.3470, Accuracy: 261/321 (81.31%)\n",
      "best_loss updated:  0.34698793292045593\n",
      "epoch= 450: Train set: Average loss: 0.2527, Accuracy: 1147/1283 (89.40%)\n",
      "Valid set: Average loss: 0.3507, Accuracy: 269/321 (83.80%)\n",
      "epoch= 460: Train set: Average loss: 0.2398, Accuracy: 1151/1283 (89.71%)\n",
      "Valid set: Average loss: 0.3317, Accuracy: 274/321 (85.36%)\n",
      "best_loss updated:  0.331682691971461\n",
      "epoch= 470: Train set: Average loss: 0.2245, Accuracy: 1155/1283 (90.02%)\n",
      "Valid set: Average loss: 0.3091, Accuracy: 286/321 (89.10%)\n",
      "best_loss updated:  0.30905527869860333\n",
      "epoch= 480: Train set: Average loss: 0.2318, Accuracy: 1154/1283 (89.95%)\n",
      "Valid set: Average loss: 0.2882, Accuracy: 283/321 (88.16%)\n",
      "best_loss updated:  0.2882485290368398\n",
      "epoch= 490: Train set: Average loss: 0.2898, Accuracy: 1167/1283 (90.96%)\n",
      "Valid set: Average loss: 0.2842, Accuracy: 284/321 (88.47%)\n",
      "best_loss updated:  0.28422678510348004\n",
      "epoch= 500: Train set: Average loss: 0.2568, Accuracy: 1166/1283 (90.88%)\n",
      "Valid set: Average loss: 0.2617, Accuracy: 276/321 (85.98%)\n",
      "best_loss updated:  0.26168113946914673\n",
      "epoch= 510: Train set: Average loss: 0.2543, Accuracy: 1145/1283 (89.24%)\n",
      "Valid set: Average loss: 0.2658, Accuracy: 283/321 (88.16%)\n",
      "epoch= 520: Train set: Average loss: 0.2739, Accuracy: 1133/1283 (88.31%)\n",
      "Valid set: Average loss: 0.3552, Accuracy: 271/321 (84.42%)\n",
      "epoch= 530: Train set: Average loss: 0.2712, Accuracy: 1163/1283 (90.65%)\n",
      "Valid set: Average loss: 0.3345, Accuracy: 269/321 (83.80%)\n",
      "epoch= 540: Train set: Average loss: 0.2512, Accuracy: 1154/1283 (89.95%)\n",
      "Valid set: Average loss: 0.2839, Accuracy: 280/321 (87.23%)\n",
      "epoch= 550: Train set: Average loss: 0.2204, Accuracy: 1168/1283 (91.04%)\n",
      "Valid set: Average loss: 0.2757, Accuracy: 279/321 (86.92%)\n",
      "epoch= 560: Train set: Average loss: 0.2483, Accuracy: 1140/1283 (88.85%)\n",
      "Valid set: Average loss: 0.3273, Accuracy: 272/321 (84.74%)\n",
      "epoch= 570: Train set: Average loss: 0.2175, Accuracy: 1162/1283 (90.57%)\n",
      "Valid set: Average loss: 0.2726, Accuracy: 283/321 (88.16%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 580: Train set: Average loss: 0.2365, Accuracy: 1163/1283 (90.65%)\n",
      "Valid set: Average loss: 0.3100, Accuracy: 275/321 (85.67%)\n",
      "epoch= 590: Train set: Average loss: 0.2204, Accuracy: 1165/1283 (90.80%)\n",
      "Valid set: Average loss: 0.2785, Accuracy: 280/321 (87.23%)\n",
      "epoch= 600: Train set: Average loss: 0.3223, Accuracy: 1156/1283 (90.10%)\n",
      "Valid set: Average loss: 0.4270, Accuracy: 265/321 (82.55%)\n",
      "epoch= 610: Train set: Average loss: 0.2626, Accuracy: 1167/1283 (90.96%)\n",
      "Valid set: Average loss: 0.2521, Accuracy: 288/321 (89.72%)\n",
      "best_loss updated:  0.25213747719923657\n",
      "epoch= 620: Train set: Average loss: 0.2487, Accuracy: 1162/1283 (90.57%)\n",
      "Valid set: Average loss: 0.2675, Accuracy: 287/321 (89.41%)\n",
      "epoch= 630: Train set: Average loss: 0.2999, Accuracy: 1156/1283 (90.10%)\n",
      "Valid set: Average loss: 0.2530, Accuracy: 284/321 (88.47%)\n",
      "epoch= 640: Train set: Average loss: 0.2175, Accuracy: 1171/1283 (91.27%)\n",
      "Valid set: Average loss: 0.2864, Accuracy: 287/321 (89.41%)\n",
      "epoch= 650: Train set: Average loss: 0.2526, Accuracy: 1187/1283 (92.52%)\n",
      "Valid set: Average loss: 0.3206, Accuracy: 279/321 (86.92%)\n",
      "epoch= 660: Train set: Average loss: 0.3353, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.2850, Accuracy: 283/321 (88.16%)\n",
      "epoch= 670: Train set: Average loss: 0.2455, Accuracy: 1175/1283 (91.58%)\n",
      "Valid set: Average loss: 0.2642, Accuracy: 286/321 (89.10%)\n",
      "epoch= 680: Train set: Average loss: 0.2826, Accuracy: 1156/1283 (90.10%)\n",
      "Valid set: Average loss: 0.3076, Accuracy: 279/321 (86.92%)\n",
      "epoch= 690: Train set: Average loss: 0.2194, Accuracy: 1175/1283 (91.58%)\n",
      "Valid set: Average loss: 0.2662, Accuracy: 287/321 (89.41%)\n",
      "epoch= 700: Train set: Average loss: 0.2350, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.3329, Accuracy: 275/321 (85.67%)\n",
      "epoch= 710: Train set: Average loss: 0.2245, Accuracy: 1159/1283 (90.34%)\n",
      "Valid set: Average loss: 0.3568, Accuracy: 271/321 (84.42%)\n",
      "epoch= 720: Train set: Average loss: 0.2194, Accuracy: 1154/1283 (89.95%)\n",
      "Valid set: Average loss: 0.2693, Accuracy: 285/321 (88.79%)\n",
      "epoch= 730: Train set: Average loss: 0.2745, Accuracy: 1170/1283 (91.19%)\n",
      "Valid set: Average loss: 0.2535, Accuracy: 292/321 (90.97%)\n",
      "epoch= 740: Train set: Average loss: 0.2961, Accuracy: 1184/1283 (92.28%)\n",
      "Valid set: Average loss: 0.3046, Accuracy: 278/321 (86.60%)\n",
      "epoch= 750: Train set: Average loss: 0.1931, Accuracy: 1177/1283 (91.74%)\n",
      "Valid set: Average loss: 0.2476, Accuracy: 292/321 (90.97%)\n",
      "best_loss updated:  0.24761938055356345\n",
      "epoch= 760: Train set: Average loss: 0.1795, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.2586, Accuracy: 292/321 (90.97%)\n",
      "epoch= 770: Train set: Average loss: 0.3254, Accuracy: 1175/1283 (91.58%)\n",
      "Valid set: Average loss: 0.2665, Accuracy: 291/321 (90.65%)\n",
      "epoch= 780: Train set: Average loss: 0.2374, Accuracy: 1159/1283 (90.34%)\n",
      "Valid set: Average loss: 0.6686, Accuracy: 253/321 (78.82%)\n",
      "epoch= 790: Train set: Average loss: 0.2283, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.2646, Accuracy: 285/321 (88.79%)\n",
      "epoch= 800: Train set: Average loss: 0.2149, Accuracy: 1175/1283 (91.58%)\n",
      "Valid set: Average loss: 0.3423, Accuracy: 274/321 (85.36%)\n",
      "epoch= 810: Train set: Average loss: 0.2462, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.2355, Accuracy: 290/321 (90.34%)\n",
      "best_loss updated:  0.23546713093916574\n",
      "epoch= 820: Train set: Average loss: 0.2778, Accuracy: 1183/1283 (92.21%)\n",
      "Valid set: Average loss: 0.2356, Accuracy: 293/321 (91.28%)\n",
      "epoch= 830: Train set: Average loss: 0.1891, Accuracy: 1178/1283 (91.82%)\n",
      "Valid set: Average loss: 0.2519, Accuracy: 289/321 (90.03%)\n",
      "epoch= 840: Train set: Average loss: 0.2493, Accuracy: 1149/1283 (89.56%)\n",
      "Valid set: Average loss: 0.2646, Accuracy: 284/321 (88.47%)\n",
      "epoch= 850: Train set: Average loss: 0.2576, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.2701, Accuracy: 287/321 (89.41%)\n",
      "epoch= 860: Train set: Average loss: 0.2147, Accuracy: 1175/1283 (91.58%)\n",
      "Valid set: Average loss: 0.2800, Accuracy: 283/321 (88.16%)\n",
      "epoch= 870: Train set: Average loss: 0.2210, Accuracy: 1201/1283 (93.61%)\n",
      "Valid set: Average loss: 0.2517, Accuracy: 290/321 (90.34%)\n",
      "epoch= 880: Train set: Average loss: 0.2102, Accuracy: 1167/1283 (90.96%)\n",
      "Valid set: Average loss: 0.3074, Accuracy: 285/321 (88.79%)\n",
      "epoch= 890: Train set: Average loss: 0.2253, Accuracy: 1171/1283 (91.27%)\n",
      "Valid set: Average loss: 0.6786, Accuracy: 255/321 (79.44%)\n",
      "epoch= 900: Train set: Average loss: 0.1777, Accuracy: 1184/1283 (92.28%)\n",
      "Valid set: Average loss: 0.2664, Accuracy: 288/321 (89.72%)\n",
      "epoch= 910: Train set: Average loss: 0.1860, Accuracy: 1189/1283 (92.67%)\n",
      "Valid set: Average loss: 0.2716, Accuracy: 289/321 (90.03%)\n",
      "epoch= 920: Train set: Average loss: 0.1852, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.2805, Accuracy: 289/321 (90.03%)\n",
      "epoch= 930: Train set: Average loss: 0.1815, Accuracy: 1187/1283 (92.52%)\n",
      "Valid set: Average loss: 0.4386, Accuracy: 272/321 (84.74%)\n",
      "epoch= 940: Train set: Average loss: 0.2079, Accuracy: 1181/1283 (92.05%)\n",
      "Valid set: Average loss: 0.3697, Accuracy: 263/321 (81.93%)\n",
      "epoch= 950: Train set: Average loss: 0.1929, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.2761, Accuracy: 287/321 (89.41%)\n",
      "epoch= 960: Train set: Average loss: 0.3031, Accuracy: 1133/1283 (88.31%)\n",
      "Valid set: Average loss: 0.2972, Accuracy: 283/321 (88.16%)\n",
      "epoch= 970: Train set: Average loss: 0.3014, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.3142, Accuracy: 281/321 (87.54%)\n",
      "epoch= 980: Train set: Average loss: 0.1757, Accuracy: 1198/1283 (93.37%)\n",
      "Valid set: Average loss: 0.2637, Accuracy: 292/321 (90.97%)\n",
      "epoch= 990: Train set: Average loss: 0.2655, Accuracy: 1186/1283 (92.44%)\n",
      "Valid set: Average loss: 0.2884, Accuracy: 287/321 (89.41%)\n",
      "epoch= 1000: Train set: Average loss: 0.2442, Accuracy: 1142/1283 (89.01%)\n",
      "Valid set: Average loss: 0.2568, Accuracy: 288/321 (89.72%)\n",
      "epoch= 1010: Train set: Average loss: 0.2115, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.3243, Accuracy: 279/321 (86.92%)\n",
      "epoch= 1020: Train set: Average loss: 0.1797, Accuracy: 1183/1283 (92.21%)\n",
      "Valid set: Average loss: 0.2652, Accuracy: 286/321 (89.10%)\n",
      "epoch= 1030: Train set: Average loss: 0.2094, Accuracy: 1197/1283 (93.30%)\n",
      "Valid set: Average loss: 0.2489, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1040: Train set: Average loss: 0.1985, Accuracy: 1178/1283 (91.82%)\n",
      "Valid set: Average loss: 0.2431, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1050: Train set: Average loss: 0.1681, Accuracy: 1189/1283 (92.67%)\n",
      "Valid set: Average loss: 0.2616, Accuracy: 295/321 (91.90%)\n",
      "epoch= 1060: Train set: Average loss: 0.2093, Accuracy: 1202/1283 (93.69%)\n",
      "Valid set: Average loss: 0.2609, Accuracy: 293/321 (91.28%)\n",
      "epoch= 1070: Train set: Average loss: 0.2773, Accuracy: 1189/1283 (92.67%)\n",
      "Valid set: Average loss: 0.2654, Accuracy: 289/321 (90.03%)\n",
      "epoch= 1080: Train set: Average loss: 0.1771, Accuracy: 1198/1283 (93.37%)\n",
      "Valid set: Average loss: 0.2751, Accuracy: 288/321 (89.72%)\n",
      "epoch= 1090: Train set: Average loss: 0.1808, Accuracy: 1197/1283 (93.30%)\n",
      "Valid set: Average loss: 0.2358, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1100: Train set: Average loss: 0.1488, Accuracy: 1199/1283 (93.45%)\n",
      "Valid set: Average loss: 0.2486, Accuracy: 287/321 (89.41%)\n",
      "epoch= 1110: Train set: Average loss: 0.2004, Accuracy: 1197/1283 (93.30%)\n",
      "Valid set: Average loss: 0.2649, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1120: Train set: Average loss: 0.1722, Accuracy: 1193/1283 (92.99%)\n",
      "Valid set: Average loss: 0.2494, Accuracy: 289/321 (90.03%)\n",
      "epoch= 1130: Train set: Average loss: 0.1698, Accuracy: 1208/1283 (94.15%)\n",
      "Valid set: Average loss: 0.2782, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1140: Train set: Average loss: 0.1760, Accuracy: 1189/1283 (92.67%)\n",
      "Valid set: Average loss: 0.2604, Accuracy: 290/321 (90.34%)\n",
      "epoch= 1150: Train set: Average loss: 0.2051, Accuracy: 1194/1283 (93.06%)\n",
      "Valid set: Average loss: 0.2655, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1160: Train set: Average loss: 0.1932, Accuracy: 1203/1283 (93.76%)\n",
      "Valid set: Average loss: 0.2615, Accuracy: 289/321 (90.03%)\n",
      "epoch= 1170: Train set: Average loss: 0.1719, Accuracy: 1206/1283 (94.00%)\n",
      "Valid set: Average loss: 0.2626, Accuracy: 290/321 (90.34%)\n",
      "epoch= 1180: Train set: Average loss: 0.1880, Accuracy: 1210/1283 (94.31%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Valid set: Average loss: 0.2539, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1190: Train set: Average loss: 0.1858, Accuracy: 1201/1283 (93.61%)\n",
      "Valid set: Average loss: 0.2566, Accuracy: 293/321 (91.28%)\n",
      "epoch= 1200: Train set: Average loss: 0.1937, Accuracy: 1205/1283 (93.92%)\n",
      "Valid set: Average loss: 0.2738, Accuracy: 287/321 (89.41%)\n",
      "epoch= 1210: Train set: Average loss: 0.1556, Accuracy: 1197/1283 (93.30%)\n",
      "Valid set: Average loss: 0.2491, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1220: Train set: Average loss: 0.1614, Accuracy: 1195/1283 (93.14%)\n",
      "Valid set: Average loss: 0.2722, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1230: Train set: Average loss: 0.1656, Accuracy: 1195/1283 (93.14%)\n",
      "Valid set: Average loss: 0.2568, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1240: Train set: Average loss: 0.1871, Accuracy: 1198/1283 (93.37%)\n",
      "Valid set: Average loss: 0.2434, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1250: Train set: Average loss: 0.1502, Accuracy: 1203/1283 (93.76%)\n",
      "Valid set: Average loss: 0.2532, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1260: Train set: Average loss: 0.1676, Accuracy: 1204/1283 (93.84%)\n",
      "Valid set: Average loss: 0.2746, Accuracy: 290/321 (90.34%)\n",
      "epoch= 1270: Train set: Average loss: 0.1659, Accuracy: 1202/1283 (93.69%)\n",
      "Valid set: Average loss: 0.2512, Accuracy: 290/321 (90.34%)\n",
      "epoch= 1280: Train set: Average loss: 0.1597, Accuracy: 1198/1283 (93.37%)\n",
      "Valid set: Average loss: 0.2680, Accuracy: 290/321 (90.34%)\n",
      "epoch= 1290: Train set: Average loss: 0.1639, Accuracy: 1191/1283 (92.83%)\n",
      "Valid set: Average loss: 0.2595, Accuracy: 288/321 (89.72%)\n",
      "epoch= 1300: Train set: Average loss: 0.1528, Accuracy: 1202/1283 (93.69%)\n",
      "Valid set: Average loss: 0.2588, Accuracy: 290/321 (90.34%)\n"
     ]
    }
   ],
   "source": [
    "model = lessFilterResNet50()\n",
    "# testing snap shot for essemble\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=0.005, weight_decay=0.001)\n",
    "\n",
    "tr = Trainer(\n",
    "        model,\n",
    "        optimizer,\n",
    "        epochs=1300,\n",
    "        milestones=[300, 1000, 1200],\n",
    "        gamma=0.2,\n",
    "        batch_size=128, \n",
    "        use_cuda=True, \n",
    "        gpu_idx=0)\n",
    "\n",
    "data_augmentation_args = {\n",
    "    'mirror': False, # not useful here\n",
    "    'rotate': True,\n",
    "    'scale': True,\n",
    "    'translation': True,\n",
    "    #'color jitter' : True,\n",
    "}\n",
    "tr.train(train, valid_df = valid, is_transfer_learning=is_transfer_learning, data_augmentation_args=data_augmentation_args)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "del model, optimizer, tr"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Cross Validation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## cross validate less filter ResNet5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EE024D32B0>, <torchvision.transforms.transforms.Lambda object at 0x000002EE024D3390>, <torchvision.transforms.transforms.ToTensor object at 0x000002EE024D31D0>]\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EE024D35C0>, <torchvision.transforms.transforms.ToTensor object at 0x000002EE024D35F8>]\n"
     ]
    },
    {
     "data": {
      "application/javascript": [
       "/* Put everything inside the global mpl namespace */\n",
       "window.mpl = {};\n",
       "\n",
       "\n",
       "mpl.get_websocket_type = function() {\n",
       "    if (typeof(WebSocket) !== 'undefined') {\n",
       "        return WebSocket;\n",
       "    } else if (typeof(MozWebSocket) !== 'undefined') {\n",
       "        return MozWebSocket;\n",
       "    } else {\n",
       "        alert('Your browser does not have WebSocket support.' +\n",
       "              'Please try Chrome, Safari or Firefox ≥ 6. ' +\n",
       "              'Firefox 4 and 5 are also supported but you ' +\n",
       "              'have to enable WebSockets in about:config.');\n",
       "    };\n",
       "}\n",
       "\n",
       "mpl.figure = function(figure_id, websocket, ondownload, parent_element) {\n",
       "    this.id = figure_id;\n",
       "\n",
       "    this.ws = websocket;\n",
       "\n",
       "    this.supports_binary = (this.ws.binaryType != undefined);\n",
       "\n",
       "    if (!this.supports_binary) {\n",
       "        var warnings = document.getElementById(\"mpl-warnings\");\n",
       "        if (warnings) {\n",
       "            warnings.style.display = 'block';\n",
       "            warnings.textContent = (\n",
       "                \"This browser does not support binary websocket messages. \" +\n",
       "                    \"Performance may be slow.\");\n",
       "        }\n",
       "    }\n",
       "\n",
       "    this.imageObj = new Image();\n",
       "\n",
       "    this.context = undefined;\n",
       "    this.message = undefined;\n",
       "    this.canvas = undefined;\n",
       "    this.rubberband_canvas = undefined;\n",
       "    this.rubberband_context = undefined;\n",
       "    this.format_dropdown = undefined;\n",
       "\n",
       "    this.image_mode = 'full';\n",
       "\n",
       "    this.root = $('<div/>');\n",
       "    this._root_extra_style(this.root)\n",
       "    this.root.attr('style', 'display: inline-block');\n",
       "\n",
       "    $(parent_element).append(this.root);\n",
       "\n",
       "    this._init_header(this);\n",
       "    this._init_canvas(this);\n",
       "    this._init_toolbar(this);\n",
       "\n",
       "    var fig = this;\n",
       "\n",
       "    this.waiting = false;\n",
       "\n",
       "    this.ws.onopen =  function () {\n",
       "            fig.send_message(\"supports_binary\", {value: fig.supports_binary});\n",
       "            fig.send_message(\"send_image_mode\", {});\n",
       "            if (mpl.ratio != 1) {\n",
       "                fig.send_message(\"set_dpi_ratio\", {'dpi_ratio': mpl.ratio});\n",
       "            }\n",
       "            fig.send_message(\"refresh\", {});\n",
       "        }\n",
       "\n",
       "    this.imageObj.onload = function() {\n",
       "            if (fig.image_mode == 'full') {\n",
       "                // Full images could contain transparency (where diff images\n",
       "                // almost always do), so we need to clear the canvas so that\n",
       "                // there is no ghosting.\n",
       "                fig.context.clearRect(0, 0, fig.canvas.width, fig.canvas.height);\n",
       "            }\n",
       "            fig.context.drawImage(fig.imageObj, 0, 0);\n",
       "        };\n",
       "\n",
       "    this.imageObj.onunload = function() {\n",
       "        this.ws.close();\n",
       "    }\n",
       "\n",
       "    this.ws.onmessage = this._make_on_message_function(this);\n",
       "\n",
       "    this.ondownload = ondownload;\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._init_header = function() {\n",
       "    var titlebar = $(\n",
       "        '<div class=\"ui-dialog-titlebar ui-widget-header ui-corner-all ' +\n",
       "        'ui-helper-clearfix\"/>');\n",
       "    var titletext = $(\n",
       "        '<div class=\"ui-dialog-title\" style=\"width: 100%; ' +\n",
       "        'text-align: center; padding: 3px;\"/>');\n",
       "    titlebar.append(titletext)\n",
       "    this.root.append(titlebar);\n",
       "    this.header = titletext[0];\n",
       "}\n",
       "\n",
       "\n",
       "\n",
       "mpl.figure.prototype._canvas_extra_style = function(canvas_div) {\n",
       "\n",
       "}\n",
       "\n",
       "\n",
       "mpl.figure.prototype._root_extra_style = function(canvas_div) {\n",
       "\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._init_canvas = function() {\n",
       "    var fig = this;\n",
       "\n",
       "    var canvas_div = $('<div/>');\n",
       "\n",
       "    canvas_div.attr('style', 'position: relative; clear: both; outline: 0');\n",
       "\n",
       "    function canvas_keyboard_event(event) {\n",
       "        return fig.key_event(event, event['data']);\n",
       "    }\n",
       "\n",
       "    canvas_div.keydown('key_press', canvas_keyboard_event);\n",
       "    canvas_div.keyup('key_release', canvas_keyboard_event);\n",
       "    this.canvas_div = canvas_div\n",
       "    this._canvas_extra_style(canvas_div)\n",
       "    this.root.append(canvas_div);\n",
       "\n",
       "    var canvas = $('<canvas/>');\n",
       "    canvas.addClass('mpl-canvas');\n",
       "    canvas.attr('style', \"left: 0; top: 0; z-index: 0; outline: 0\")\n",
       "\n",
       "    this.canvas = canvas[0];\n",
       "    this.context = canvas[0].getContext(\"2d\");\n",
       "\n",
       "    var backingStore = this.context.backingStorePixelRatio ||\n",
       "\tthis.context.webkitBackingStorePixelRatio ||\n",
       "\tthis.context.mozBackingStorePixelRatio ||\n",
       "\tthis.context.msBackingStorePixelRatio ||\n",
       "\tthis.context.oBackingStorePixelRatio ||\n",
       "\tthis.context.backingStorePixelRatio || 1;\n",
       "\n",
       "    mpl.ratio = (window.devicePixelRatio || 1) / backingStore;\n",
       "\n",
       "    var rubberband = $('<canvas/>');\n",
       "    rubberband.attr('style', \"position: absolute; left: 0; top: 0; z-index: 1;\")\n",
       "\n",
       "    var pass_mouse_events = true;\n",
       "\n",
       "    canvas_div.resizable({\n",
       "        start: function(event, ui) {\n",
       "            pass_mouse_events = false;\n",
       "        },\n",
       "        resize: function(event, ui) {\n",
       "            fig.request_resize(ui.size.width, ui.size.height);\n",
       "        },\n",
       "        stop: function(event, ui) {\n",
       "            pass_mouse_events = true;\n",
       "            fig.request_resize(ui.size.width, ui.size.height);\n",
       "        },\n",
       "    });\n",
       "\n",
       "    function mouse_event_fn(event) {\n",
       "        if (pass_mouse_events)\n",
       "            return fig.mouse_event(event, event['data']);\n",
       "    }\n",
       "\n",
       "    rubberband.mousedown('button_press', mouse_event_fn);\n",
       "    rubberband.mouseup('button_release', mouse_event_fn);\n",
       "    // Throttle sequential mouse events to 1 every 20ms.\n",
       "    rubberband.mousemove('motion_notify', mouse_event_fn);\n",
       "\n",
       "    rubberband.mouseenter('figure_enter', mouse_event_fn);\n",
       "    rubberband.mouseleave('figure_leave', mouse_event_fn);\n",
       "\n",
       "    canvas_div.on(\"wheel\", function (event) {\n",
       "        event = event.originalEvent;\n",
       "        event['data'] = 'scroll'\n",
       "        if (event.deltaY < 0) {\n",
       "            event.step = 1;\n",
       "        } else {\n",
       "            event.step = -1;\n",
       "        }\n",
       "        mouse_event_fn(event);\n",
       "    });\n",
       "\n",
       "    canvas_div.append(canvas);\n",
       "    canvas_div.append(rubberband);\n",
       "\n",
       "    this.rubberband = rubberband;\n",
       "    this.rubberband_canvas = rubberband[0];\n",
       "    this.rubberband_context = rubberband[0].getContext(\"2d\");\n",
       "    this.rubberband_context.strokeStyle = \"#000000\";\n",
       "\n",
       "    this._resize_canvas = function(width, height) {\n",
       "        // Keep the size of the canvas, canvas container, and rubber band\n",
       "        // canvas in synch.\n",
       "        canvas_div.css('width', width)\n",
       "        canvas_div.css('height', height)\n",
       "\n",
       "        canvas.attr('width', width * mpl.ratio);\n",
       "        canvas.attr('height', height * mpl.ratio);\n",
       "        canvas.attr('style', 'width: ' + width + 'px; height: ' + height + 'px;');\n",
       "\n",
       "        rubberband.attr('width', width);\n",
       "        rubberband.attr('height', height);\n",
       "    }\n",
       "\n",
       "    // Set the figure to an initial 600x600px, this will subsequently be updated\n",
       "    // upon first draw.\n",
       "    this._resize_canvas(600, 600);\n",
       "\n",
       "    // Disable right mouse context menu.\n",
       "    $(this.rubberband_canvas).bind(\"contextmenu\",function(e){\n",
       "        return false;\n",
       "    });\n",
       "\n",
       "    function set_focus () {\n",
       "        canvas.focus();\n",
       "        canvas_div.focus();\n",
       "    }\n",
       "\n",
       "    window.setTimeout(set_focus, 100);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._init_toolbar = function() {\n",
       "    var fig = this;\n",
       "\n",
       "    var nav_element = $('<div/>')\n",
       "    nav_element.attr('style', 'width: 100%');\n",
       "    this.root.append(nav_element);\n",
       "\n",
       "    // Define a callback function for later on.\n",
       "    function toolbar_event(event) {\n",
       "        return fig.toolbar_button_onclick(event['data']);\n",
       "    }\n",
       "    function toolbar_mouse_event(event) {\n",
       "        return fig.toolbar_button_onmouseover(event['data']);\n",
       "    }\n",
       "\n",
       "    for(var toolbar_ind in mpl.toolbar_items) {\n",
       "        var name = mpl.toolbar_items[toolbar_ind][0];\n",
       "        var tooltip = mpl.toolbar_items[toolbar_ind][1];\n",
       "        var image = mpl.toolbar_items[toolbar_ind][2];\n",
       "        var method_name = mpl.toolbar_items[toolbar_ind][3];\n",
       "\n",
       "        if (!name) {\n",
       "            // put a spacer in here.\n",
       "            continue;\n",
       "        }\n",
       "        var button = $('<button/>');\n",
       "        button.addClass('ui-button ui-widget ui-state-default ui-corner-all ' +\n",
       "                        'ui-button-icon-only');\n",
       "        button.attr('role', 'button');\n",
       "        button.attr('aria-disabled', 'false');\n",
       "        button.click(method_name, toolbar_event);\n",
       "        button.mouseover(tooltip, toolbar_mouse_event);\n",
       "\n",
       "        var icon_img = $('<span/>');\n",
       "        icon_img.addClass('ui-button-icon-primary ui-icon');\n",
       "        icon_img.addClass(image);\n",
       "        icon_img.addClass('ui-corner-all');\n",
       "\n",
       "        var tooltip_span = $('<span/>');\n",
       "        tooltip_span.addClass('ui-button-text');\n",
       "        tooltip_span.html(tooltip);\n",
       "\n",
       "        button.append(icon_img);\n",
       "        button.append(tooltip_span);\n",
       "\n",
       "        nav_element.append(button);\n",
       "    }\n",
       "\n",
       "    var fmt_picker_span = $('<span/>');\n",
       "\n",
       "    var fmt_picker = $('<select/>');\n",
       "    fmt_picker.addClass('mpl-toolbar-option ui-widget ui-widget-content');\n",
       "    fmt_picker_span.append(fmt_picker);\n",
       "    nav_element.append(fmt_picker_span);\n",
       "    this.format_dropdown = fmt_picker[0];\n",
       "\n",
       "    for (var ind in mpl.extensions) {\n",
       "        var fmt = mpl.extensions[ind];\n",
       "        var option = $(\n",
       "            '<option/>', {selected: fmt === mpl.default_extension}).html(fmt);\n",
       "        fmt_picker.append(option)\n",
       "    }\n",
       "\n",
       "    // Add hover states to the ui-buttons\n",
       "    $( \".ui-button\" ).hover(\n",
       "        function() { $(this).addClass(\"ui-state-hover\");},\n",
       "        function() { $(this).removeClass(\"ui-state-hover\");}\n",
       "    );\n",
       "\n",
       "    var status_bar = $('<span class=\"mpl-message\"/>');\n",
       "    nav_element.append(status_bar);\n",
       "    this.message = status_bar[0];\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.request_resize = function(x_pixels, y_pixels) {\n",
       "    // Request matplotlib to resize the figure. Matplotlib will then trigger a resize in the client,\n",
       "    // which will in turn request a refresh of the image.\n",
       "    this.send_message('resize', {'width': x_pixels, 'height': y_pixels});\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.send_message = function(type, properties) {\n",
       "    properties['type'] = type;\n",
       "    properties['figure_id'] = this.id;\n",
       "    this.ws.send(JSON.stringify(properties));\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.send_draw_message = function() {\n",
       "    if (!this.waiting) {\n",
       "        this.waiting = true;\n",
       "        this.ws.send(JSON.stringify({type: \"draw\", figure_id: this.id}));\n",
       "    }\n",
       "}\n",
       "\n",
       "\n",
       "mpl.figure.prototype.handle_save = function(fig, msg) {\n",
       "    var format_dropdown = fig.format_dropdown;\n",
       "    var format = format_dropdown.options[format_dropdown.selectedIndex].value;\n",
       "    fig.ondownload(fig, format);\n",
       "}\n",
       "\n",
       "\n",
       "mpl.figure.prototype.handle_resize = function(fig, msg) {\n",
       "    var size = msg['size'];\n",
       "    if (size[0] != fig.canvas.width || size[1] != fig.canvas.height) {\n",
       "        fig._resize_canvas(size[0], size[1]);\n",
       "        fig.send_message(\"refresh\", {});\n",
       "    };\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_rubberband = function(fig, msg) {\n",
       "    var x0 = msg['x0'] / mpl.ratio;\n",
       "    var y0 = (fig.canvas.height - msg['y0']) / mpl.ratio;\n",
       "    var x1 = msg['x1'] / mpl.ratio;\n",
       "    var y1 = (fig.canvas.height - msg['y1']) / mpl.ratio;\n",
       "    x0 = Math.floor(x0) + 0.5;\n",
       "    y0 = Math.floor(y0) + 0.5;\n",
       "    x1 = Math.floor(x1) + 0.5;\n",
       "    y1 = Math.floor(y1) + 0.5;\n",
       "    var min_x = Math.min(x0, x1);\n",
       "    var min_y = Math.min(y0, y1);\n",
       "    var width = Math.abs(x1 - x0);\n",
       "    var height = Math.abs(y1 - y0);\n",
       "\n",
       "    fig.rubberband_context.clearRect(\n",
       "        0, 0, fig.canvas.width, fig.canvas.height);\n",
       "\n",
       "    fig.rubberband_context.strokeRect(min_x, min_y, width, height);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_figure_label = function(fig, msg) {\n",
       "    // Updates the figure title.\n",
       "    fig.header.textContent = msg['label'];\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_cursor = function(fig, msg) {\n",
       "    var cursor = msg['cursor'];\n",
       "    switch(cursor)\n",
       "    {\n",
       "    case 0:\n",
       "        cursor = 'pointer';\n",
       "        break;\n",
       "    case 1:\n",
       "        cursor = 'default';\n",
       "        break;\n",
       "    case 2:\n",
       "        cursor = 'crosshair';\n",
       "        break;\n",
       "    case 3:\n",
       "        cursor = 'move';\n",
       "        break;\n",
       "    }\n",
       "    fig.rubberband_canvas.style.cursor = cursor;\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_message = function(fig, msg) {\n",
       "    fig.message.textContent = msg['message'];\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_draw = function(fig, msg) {\n",
       "    // Request the server to send over a new figure.\n",
       "    fig.send_draw_message();\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_image_mode = function(fig, msg) {\n",
       "    fig.image_mode = msg['mode'];\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.updated_canvas_event = function() {\n",
       "    // Called whenever the canvas gets updated.\n",
       "    this.send_message(\"ack\", {});\n",
       "}\n",
       "\n",
       "// A function to construct a web socket function for onmessage handling.\n",
       "// Called in the figure constructor.\n",
       "mpl.figure.prototype._make_on_message_function = function(fig) {\n",
       "    return function socket_on_message(evt) {\n",
       "        if (evt.data instanceof Blob) {\n",
       "            /* FIXME: We get \"Resource interpreted as Image but\n",
       "             * transferred with MIME type text/plain:\" errors on\n",
       "             * Chrome.  But how to set the MIME type?  It doesn't seem\n",
       "             * to be part of the websocket stream */\n",
       "            evt.data.type = \"image/png\";\n",
       "\n",
       "            /* Free the memory for the previous frames */\n",
       "            if (fig.imageObj.src) {\n",
       "                (window.URL || window.webkitURL).revokeObjectURL(\n",
       "                    fig.imageObj.src);\n",
       "            }\n",
       "\n",
       "            fig.imageObj.src = (window.URL || window.webkitURL).createObjectURL(\n",
       "                evt.data);\n",
       "            fig.updated_canvas_event();\n",
       "            fig.waiting = false;\n",
       "            return;\n",
       "        }\n",
       "        else if (typeof evt.data === 'string' && evt.data.slice(0, 21) == \"data:image/png;base64\") {\n",
       "            fig.imageObj.src = evt.data;\n",
       "            fig.updated_canvas_event();\n",
       "            fig.waiting = false;\n",
       "            return;\n",
       "        }\n",
       "\n",
       "        var msg = JSON.parse(evt.data);\n",
       "        var msg_type = msg['type'];\n",
       "\n",
       "        // Call the  \"handle_{type}\" callback, which takes\n",
       "        // the figure and JSON message as its only arguments.\n",
       "        try {\n",
       "            var callback = fig[\"handle_\" + msg_type];\n",
       "        } catch (e) {\n",
       "            console.log(\"No handler for the '\" + msg_type + \"' message type: \", msg);\n",
       "            return;\n",
       "        }\n",
       "\n",
       "        if (callback) {\n",
       "            try {\n",
       "                // console.log(\"Handling '\" + msg_type + \"' message: \", msg);\n",
       "                callback(fig, msg);\n",
       "            } catch (e) {\n",
       "                console.log(\"Exception inside the 'handler_\" + msg_type + \"' callback:\", e, e.stack, msg);\n",
       "            }\n",
       "        }\n",
       "    };\n",
       "}\n",
       "\n",
       "// from http://stackoverflow.com/questions/1114465/getting-mouse-location-in-canvas\n",
       "mpl.findpos = function(e) {\n",
       "    //this section is from http://www.quirksmode.org/js/events_properties.html\n",
       "    var targ;\n",
       "    if (!e)\n",
       "        e = window.event;\n",
       "    if (e.target)\n",
       "        targ = e.target;\n",
       "    else if (e.srcElement)\n",
       "        targ = e.srcElement;\n",
       "    if (targ.nodeType == 3) // defeat Safari bug\n",
       "        targ = targ.parentNode;\n",
       "\n",
       "    // jQuery normalizes the pageX and pageY\n",
       "    // pageX,Y are the mouse positions relative to the document\n",
       "    // offset() returns the position of the element relative to the document\n",
       "    var x = e.pageX - $(targ).offset().left;\n",
       "    var y = e.pageY - $(targ).offset().top;\n",
       "\n",
       "    return {\"x\": x, \"y\": y};\n",
       "};\n",
       "\n",
       "/*\n",
       " * return a copy of an object with only non-object keys\n",
       " * we need this to avoid circular references\n",
       " * http://stackoverflow.com/a/24161582/3208463\n",
       " */\n",
       "function simpleKeys (original) {\n",
       "  return Object.keys(original).reduce(function (obj, key) {\n",
       "    if (typeof original[key] !== 'object')\n",
       "        obj[key] = original[key]\n",
       "    return obj;\n",
       "  }, {});\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.mouse_event = function(event, name) {\n",
       "    var canvas_pos = mpl.findpos(event)\n",
       "\n",
       "    if (name === 'button_press')\n",
       "    {\n",
       "        this.canvas.focus();\n",
       "        this.canvas_div.focus();\n",
       "    }\n",
       "\n",
       "    var x = canvas_pos.x * mpl.ratio;\n",
       "    var y = canvas_pos.y * mpl.ratio;\n",
       "\n",
       "    this.send_message(name, {x: x, y: y, button: event.button,\n",
       "                             step: event.step,\n",
       "                             guiEvent: simpleKeys(event)});\n",
       "\n",
       "    /* This prevents the web browser from automatically changing to\n",
       "     * the text insertion cursor when the button is pressed.  We want\n",
       "     * to control all of the cursor setting manually through the\n",
       "     * 'cursor' event from matplotlib */\n",
       "    event.preventDefault();\n",
       "    return false;\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._key_event_extra = function(event, name) {\n",
       "    // Handle any extra behaviour associated with a key event\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.key_event = function(event, name) {\n",
       "\n",
       "    // Prevent repeat events\n",
       "    if (name == 'key_press')\n",
       "    {\n",
       "        if (event.which === this._key)\n",
       "            return;\n",
       "        else\n",
       "            this._key = event.which;\n",
       "    }\n",
       "    if (name == 'key_release')\n",
       "        this._key = null;\n",
       "\n",
       "    var value = '';\n",
       "    if (event.ctrlKey && event.which != 17)\n",
       "        value += \"ctrl+\";\n",
       "    if (event.altKey && event.which != 18)\n",
       "        value += \"alt+\";\n",
       "    if (event.shiftKey && event.which != 16)\n",
       "        value += \"shift+\";\n",
       "\n",
       "    value += 'k';\n",
       "    value += event.which.toString();\n",
       "\n",
       "    this._key_event_extra(event, name);\n",
       "\n",
       "    this.send_message(name, {key: value,\n",
       "                             guiEvent: simpleKeys(event)});\n",
       "    return false;\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.toolbar_button_onclick = function(name) {\n",
       "    if (name == 'download') {\n",
       "        this.handle_save(this, null);\n",
       "    } else {\n",
       "        this.send_message(\"toolbar_button\", {name: name});\n",
       "    }\n",
       "};\n",
       "\n",
       "mpl.figure.prototype.toolbar_button_onmouseover = function(tooltip) {\n",
       "    this.message.textContent = tooltip;\n",
       "};\n",
       "mpl.toolbar_items = [[\"Home\", \"Reset original view\", \"fa fa-home icon-home\", \"home\"], [\"Back\", \"Back to  previous view\", \"fa fa-arrow-left icon-arrow-left\", \"back\"], [\"Forward\", \"Forward to next view\", \"fa fa-arrow-right icon-arrow-right\", \"forward\"], [\"\", \"\", \"\", \"\"], [\"Pan\", \"Pan axes with left mouse, zoom with right\", \"fa fa-arrows icon-move\", \"pan\"], [\"Zoom\", \"Zoom to rectangle\", \"fa fa-square-o icon-check-empty\", \"zoom\"], [\"\", \"\", \"\", \"\"], [\"Download\", \"Download plot\", \"fa fa-floppy-o icon-save\", \"download\"]];\n",
       "\n",
       "mpl.extensions = [\"eps\", \"jpeg\", \"pdf\", \"png\", \"ps\", \"raw\", \"svg\", \"tif\"];\n",
       "\n",
       "mpl.default_extension = \"png\";var comm_websocket_adapter = function(comm) {\n",
       "    // Create a \"websocket\"-like object which calls the given IPython comm\n",
       "    // object with the appropriate methods. Currently this is a non binary\n",
       "    // socket, so there is still some room for performance tuning.\n",
       "    var ws = {};\n",
       "\n",
       "    ws.close = function() {\n",
       "        comm.close()\n",
       "    };\n",
       "    ws.send = function(m) {\n",
       "        //console.log('sending', m);\n",
       "        comm.send(m);\n",
       "    };\n",
       "    // Register the callback with on_msg.\n",
       "    comm.on_msg(function(msg) {\n",
       "        //console.log('receiving', msg['content']['data'], msg);\n",
       "        // Pass the mpl event to the overriden (by mpl) onmessage function.\n",
       "        ws.onmessage(msg['content']['data'])\n",
       "    });\n",
       "    return ws;\n",
       "}\n",
       "\n",
       "mpl.mpl_figure_comm = function(comm, msg) {\n",
       "    // This is the function which gets called when the mpl process\n",
       "    // starts-up an IPython Comm through the \"matplotlib\" channel.\n",
       "\n",
       "    var id = msg.content.data.id;\n",
       "    // Get hold of the div created by the display call when the Comm\n",
       "    // socket was opened in Python.\n",
       "    var element = $(\"#\" + id);\n",
       "    var ws_proxy = comm_websocket_adapter(comm)\n",
       "\n",
       "    function ondownload(figure, format) {\n",
       "        window.open(figure.imageObj.src);\n",
       "    }\n",
       "\n",
       "    var fig = new mpl.figure(id, ws_proxy,\n",
       "                           ondownload,\n",
       "                           element.get(0));\n",
       "\n",
       "    // Call onopen now - mpl needs it, as it is assuming we've passed it a real\n",
       "    // web socket which is closed, not our websocket->open comm proxy.\n",
       "    ws_proxy.onopen();\n",
       "\n",
       "    fig.parent_element = element.get(0);\n",
       "    fig.cell_info = mpl.find_output_cell(\"<div id='\" + id + \"'></div>\");\n",
       "    if (!fig.cell_info) {\n",
       "        console.error(\"Failed to find cell for figure\", id, fig);\n",
       "        return;\n",
       "    }\n",
       "\n",
       "    var output_index = fig.cell_info[2]\n",
       "    var cell = fig.cell_info[0];\n",
       "\n",
       "};\n",
       "\n",
       "mpl.figure.prototype.handle_close = function(fig, msg) {\n",
       "    var width = fig.canvas.width/mpl.ratio\n",
       "    fig.root.unbind('remove')\n",
       "\n",
       "    // Update the output cell to use the data from the current canvas.\n",
       "    fig.push_to_output();\n",
       "    var dataURL = fig.canvas.toDataURL();\n",
       "    // Re-enable the keyboard manager in IPython - without this line, in FF,\n",
       "    // the notebook keyboard shortcuts fail.\n",
       "    IPython.keyboard_manager.enable()\n",
       "    $(fig.parent_element).html('<img src=\"' + dataURL + '\" width=\"' + width + '\">');\n",
       "    fig.close_ws(fig, msg);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.close_ws = function(fig, msg){\n",
       "    fig.send_message('closing', msg);\n",
       "    // fig.ws.close()\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.push_to_output = function(remove_interactive) {\n",
       "    // Turn the data on the canvas into data in the output cell.\n",
       "    var width = this.canvas.width/mpl.ratio\n",
       "    var dataURL = this.canvas.toDataURL();\n",
       "    this.cell_info[1]['text/html'] = '<img src=\"' + dataURL + '\" width=\"' + width + '\">';\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.updated_canvas_event = function() {\n",
       "    // Tell IPython that the notebook contents must change.\n",
       "    IPython.notebook.set_dirty(true);\n",
       "    this.send_message(\"ack\", {});\n",
       "    var fig = this;\n",
       "    // Wait a second, then push the new image to the DOM so\n",
       "    // that it is saved nicely (might be nice to debounce this).\n",
       "    setTimeout(function () { fig.push_to_output() }, 1000);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._init_toolbar = function() {\n",
       "    var fig = this;\n",
       "\n",
       "    var nav_element = $('<div/>')\n",
       "    nav_element.attr('style', 'width: 100%');\n",
       "    this.root.append(nav_element);\n",
       "\n",
       "    // Define a callback function for later on.\n",
       "    function toolbar_event(event) {\n",
       "        return fig.toolbar_button_onclick(event['data']);\n",
       "    }\n",
       "    function toolbar_mouse_event(event) {\n",
       "        return fig.toolbar_button_onmouseover(event['data']);\n",
       "    }\n",
       "\n",
       "    for(var toolbar_ind in mpl.toolbar_items){\n",
       "        var name = mpl.toolbar_items[toolbar_ind][0];\n",
       "        var tooltip = mpl.toolbar_items[toolbar_ind][1];\n",
       "        var image = mpl.toolbar_items[toolbar_ind][2];\n",
       "        var method_name = mpl.toolbar_items[toolbar_ind][3];\n",
       "\n",
       "        if (!name) { continue; };\n",
       "\n",
       "        var button = $('<button class=\"btn btn-default\" href=\"#\" title=\"' + name + '\"><i class=\"fa ' + image + ' fa-lg\"></i></button>');\n",
       "        button.click(method_name, toolbar_event);\n",
       "        button.mouseover(tooltip, toolbar_mouse_event);\n",
       "        nav_element.append(button);\n",
       "    }\n",
       "\n",
       "    // Add the status bar.\n",
       "    var status_bar = $('<span class=\"mpl-message\" style=\"text-align:right; float: right;\"/>');\n",
       "    nav_element.append(status_bar);\n",
       "    this.message = status_bar[0];\n",
       "\n",
       "    // Add the close button to the window.\n",
       "    var buttongrp = $('<div class=\"btn-group inline pull-right\"></div>');\n",
       "    var button = $('<button class=\"btn btn-mini btn-primary\" href=\"#\" title=\"Stop Interaction\"><i class=\"fa fa-power-off icon-remove icon-large\"></i></button>');\n",
       "    button.click(function (evt) { fig.handle_close(fig, {}); } );\n",
       "    button.mouseover('Stop Interaction', toolbar_mouse_event);\n",
       "    buttongrp.append(button);\n",
       "    var titlebar = this.root.find($('.ui-dialog-titlebar'));\n",
       "    titlebar.prepend(buttongrp);\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._root_extra_style = function(el){\n",
       "    var fig = this\n",
       "    el.on(\"remove\", function(){\n",
       "\tfig.close_ws(fig, {});\n",
       "    });\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._canvas_extra_style = function(el){\n",
       "    // this is important to make the div 'focusable\n",
       "    el.attr('tabindex', 0)\n",
       "    // reach out to IPython and tell the keyboard manager to turn it's self\n",
       "    // off when our div gets focus\n",
       "\n",
       "    // location in version 3\n",
       "    if (IPython.notebook.keyboard_manager) {\n",
       "        IPython.notebook.keyboard_manager.register_events(el);\n",
       "    }\n",
       "    else {\n",
       "        // location in version 2\n",
       "        IPython.keyboard_manager.register_events(el);\n",
       "    }\n",
       "\n",
       "}\n",
       "\n",
       "mpl.figure.prototype._key_event_extra = function(event, name) {\n",
       "    var manager = IPython.notebook.keyboard_manager;\n",
       "    if (!manager)\n",
       "        manager = IPython.keyboard_manager;\n",
       "\n",
       "    // Check for shift+enter\n",
       "    if (event.shiftKey && event.which == 13) {\n",
       "        this.canvas_div.blur();\n",
       "        // select the cell after this one\n",
       "        var index = IPython.notebook.find_cell_index(this.cell_info[0]);\n",
       "        IPython.notebook.select(index + 1);\n",
       "    }\n",
       "}\n",
       "\n",
       "mpl.figure.prototype.handle_save = function(fig, msg) {\n",
       "    fig.ondownload(fig, null);\n",
       "}\n",
       "\n",
       "\n",
       "mpl.find_output_cell = function(html_output) {\n",
       "    // Return the cell and output element which can be found *uniquely* in the notebook.\n",
       "    // Note - this is a bit hacky, but it is done because the \"notebook_saving.Notebook\"\n",
       "    // IPython event is triggered only after the cells have been serialised, which for\n",
       "    // our purposes (turning an active figure into a static one), is too late.\n",
       "    var cells = IPython.notebook.get_cells();\n",
       "    var ncells = cells.length;\n",
       "    for (var i=0; i<ncells; i++) {\n",
       "        var cell = cells[i];\n",
       "        if (cell.cell_type === 'code'){\n",
       "            for (var j=0; j<cell.output_area.outputs.length; j++) {\n",
       "                var data = cell.output_area.outputs[j];\n",
       "                if (data.data) {\n",
       "                    // IPython >= 3 moved mimebundle to data attribute of output\n",
       "                    data = data.data;\n",
       "                }\n",
       "                if (data['text/html'] == html_output) {\n",
       "                    return [cell, data, j];\n",
       "                }\n",
       "            }\n",
       "        }\n",
       "    }\n",
       "}\n",
       "\n",
       "// Register the function which deals with the matplotlib target/channel.\n",
       "// The kernel may be null if the page has been refreshed.\n",
       "if (IPython.notebook.kernel != null) {\n",
       "    IPython.notebook.kernel.comm_manager.register_target('matplotlib', mpl.mpl_figure_comm);\n",
       "}\n"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<img src=\"data:image/png;base64,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\" width=\"1000\">"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 10: Train set: Average loss: 1.2111, Accuracy: 730/1283 (56.90%)\n",
      "Valid set: Average loss: 0.7834, Accuracy: 169/321 (52.65%)\n",
      "best_loss updated:  0.7834033767382304\n",
      "epoch= 20: Train set: Average loss: 0.5754, Accuracy: 820/1283 (63.91%)\n",
      "Valid set: Average loss: 0.6970, Accuracy: 162/321 (50.47%)\n",
      "best_loss updated:  0.6969930728276571\n",
      "epoch= 30: Train set: Average loss: 0.6560, Accuracy: 861/1283 (67.11%)\n",
      "Valid set: Average loss: 0.7040, Accuracy: 153/321 (47.66%)\n",
      "epoch= 40: Train set: Average loss: 0.5064, Accuracy: 974/1283 (75.92%)\n",
      "Valid set: Average loss: 0.5317, Accuracy: 251/321 (78.19%)\n",
      "best_loss updated:  0.531653900941213\n",
      "epoch= 50: Train set: Average loss: 0.5066, Accuracy: 951/1283 (74.12%)\n",
      "Valid set: Average loss: 0.5918, Accuracy: 229/321 (71.34%)\n",
      "epoch= 60: Train set: Average loss: 0.5246, Accuracy: 910/1283 (70.93%)\n",
      "Valid set: Average loss: 0.5434, Accuracy: 186/321 (57.94%)\n",
      "epoch= 70: Train set: Average loss: 0.4585, Accuracy: 1039/1283 (80.98%)\n",
      "Valid set: Average loss: 0.4334, Accuracy: 264/321 (82.24%)\n",
      "best_loss updated:  0.4334203203519185\n",
      "epoch= 80: Train set: Average loss: 0.5480, Accuracy: 945/1283 (73.66%)\n",
      "Valid set: Average loss: 0.5399, Accuracy: 243/321 (75.70%)\n",
      "epoch= 90: Train set: Average loss: 0.4793, Accuracy: 1008/1283 (78.57%)\n",
      "Valid set: Average loss: 0.4563, Accuracy: 256/321 (79.75%)\n",
      "epoch= 100: Train set: Average loss: 0.4908, Accuracy: 970/1283 (75.60%)\n",
      "Valid set: Average loss: 1.9010, Accuracy: 172/321 (53.58%)\n",
      "epoch= 110: Train set: Average loss: 0.5437, Accuracy: 1000/1283 (77.94%)\n",
      "Valid set: Average loss: 0.5557, Accuracy: 234/321 (72.90%)\n",
      "epoch= 120: Train set: Average loss: 0.5145, Accuracy: 1042/1283 (81.22%)\n",
      "Valid set: Average loss: 0.5285, Accuracy: 186/321 (57.94%)\n",
      "epoch= 130: Train set: Average loss: 0.4453, Accuracy: 1042/1283 (81.22%)\n",
      "Valid set: Average loss: 0.4977, Accuracy: 249/321 (77.57%)\n",
      "epoch= 140: Train set: Average loss: 0.4507, Accuracy: 1015/1283 (79.11%)\n",
      "Valid set: Average loss: 0.4320, Accuracy: 261/321 (81.31%)\n",
      "best_loss updated:  0.4320499201615651\n",
      "epoch= 150: Train set: Average loss: 0.4912, Accuracy: 969/1283 (75.53%)\n",
      "Valid set: Average loss: 3.8429, Accuracy: 169/321 (52.65%)\n",
      "epoch= 160: Train set: Average loss: 0.4855, Accuracy: 1027/1283 (80.05%)\n",
      "Valid set: Average loss: 0.4990, Accuracy: 251/321 (78.19%)\n",
      "epoch= 170: Train set: Average loss: 0.4282, Accuracy: 1052/1283 (82.00%)\n",
      "Valid set: Average loss: 0.4005, Accuracy: 269/321 (83.80%)\n",
      "best_loss updated:  0.40051652987798053\n",
      "epoch= 180: Train set: Average loss: 0.3929, Accuracy: 1051/1283 (81.92%)\n",
      "Valid set: Average loss: 0.4315, Accuracy: 263/321 (81.93%)\n",
      "epoch= 190: Train set: Average loss: 0.4306, Accuracy: 1005/1283 (78.33%)\n",
      "Valid set: Average loss: 0.4342, Accuracy: 256/321 (79.75%)\n",
      "epoch= 200: Train set: Average loss: 0.4032, Accuracy: 1029/1283 (80.20%)\n",
      "Valid set: Average loss: 0.4630, Accuracy: 263/321 (81.93%)\n",
      "epoch= 210: Train set: Average loss: 0.4036, Accuracy: 1029/1283 (80.20%)\n",
      "Valid set: Average loss: 0.4238, Accuracy: 262/321 (81.62%)\n",
      "epoch= 220: Train set: Average loss: 0.5749, Accuracy: 992/1283 (77.32%)\n",
      "Valid set: Average loss: 0.8848, Accuracy: 189/321 (58.88%)\n",
      "epoch= 230: Train set: Average loss: 0.4880, Accuracy: 1035/1283 (80.67%)\n",
      "Valid set: Average loss: 0.4853, Accuracy: 250/321 (77.88%)\n",
      "epoch= 240: Train set: Average loss: 0.5592, Accuracy: 1022/1283 (79.66%)\n",
      "Valid set: Average loss: 0.5862, Accuracy: 232/321 (72.27%)\n",
      "epoch= 250: Train set: Average loss: 0.4049, Accuracy: 1041/1283 (81.14%)\n",
      "Valid set: Average loss: 0.4490, Accuracy: 261/321 (81.31%)\n",
      "epoch= 260: Train set: Average loss: 0.3953, Accuracy: 1056/1283 (82.31%)\n",
      "Valid set: Average loss: 0.3780, Accuracy: 271/321 (84.42%)\n",
      "best_loss updated:  0.3779587248961131\n",
      "epoch= 270: Train set: Average loss: 0.3924, Accuracy: 1052/1283 (82.00%)\n",
      "Valid set: Average loss: 2.2888, Accuracy: 189/321 (58.88%)\n",
      "epoch= 280: Train set: Average loss: 0.4184, Accuracy: 1040/1283 (81.06%)\n",
      "Valid set: Average loss: 0.4115, Accuracy: 267/321 (83.18%)\n",
      "epoch= 290: Train set: Average loss: 0.4909, Accuracy: 976/1283 (76.07%)\n",
      "Valid set: Average loss: 0.5476, Accuracy: 245/321 (76.32%)\n",
      "epoch= 300: Train set: Average loss: 0.4624, Accuracy: 1016/1283 (79.19%)\n",
      "Valid set: Average loss: 0.4458, Accuracy: 262/321 (81.62%)\n",
      "epoch= 310: Train set: Average loss: 0.3979, Accuracy: 1045/1283 (81.45%)\n",
      "Valid set: Average loss: 0.4159, Accuracy: 269/321 (83.80%)\n",
      "epoch= 320: Train set: Average loss: 0.4983, Accuracy: 1070/1283 (83.40%)\n",
      "Valid set: Average loss: 0.4087, Accuracy: 264/321 (82.24%)\n",
      "epoch= 330: Train set: Average loss: 0.4000, Accuracy: 1049/1283 (81.76%)\n",
      "Valid set: Average loss: 0.4314, Accuracy: 259/321 (80.69%)\n",
      "epoch= 340: Train set: Average loss: 0.3987, Accuracy: 1055/1283 (82.23%)\n",
      "Valid set: Average loss: 0.3868, Accuracy: 269/321 (83.80%)\n",
      "epoch= 350: Train set: Average loss: 0.4401, Accuracy: 1080/1283 (84.18%)\n",
      "Valid set: Average loss: 0.3674, Accuracy: 271/321 (84.42%)\n",
      "best_loss updated:  0.36741729577382404\n",
      "epoch= 360: Train set: Average loss: 0.3815, Accuracy: 1060/1283 (82.62%)\n",
      "Valid set: Average loss: 0.4242, Accuracy: 261/321 (81.31%)\n",
      "epoch= 370: Train set: Average loss: 0.4290, Accuracy: 1059/1283 (82.54%)\n",
      "Valid set: Average loss: 0.3795, Accuracy: 272/321 (84.74%)\n",
      "epoch= 380: Train set: Average loss: 0.5061, Accuracy: 1081/1283 (84.26%)\n",
      "Valid set: Average loss: 0.3557, Accuracy: 274/321 (85.36%)\n",
      "best_loss updated:  0.3557131588459015\n",
      "epoch= 390: Train set: Average loss: 0.4240, Accuracy: 1072/1283 (83.55%)\n",
      "Valid set: Average loss: 0.3674, Accuracy: 274/321 (85.36%)\n",
      "epoch= 400: Train set: Average loss: 0.3946, Accuracy: 1080/1283 (84.18%)\n",
      "Valid set: Average loss: 0.3556, Accuracy: 278/321 (86.60%)\n",
      "best_loss updated:  0.35555235544840497\n",
      "epoch= 410: Train set: Average loss: 0.4065, Accuracy: 1075/1283 (83.79%)\n",
      "Valid set: Average loss: 0.3825, Accuracy: 267/321 (83.18%)\n",
      "epoch= 420: Train set: Average loss: 0.4779, Accuracy: 1087/1283 (84.72%)\n",
      "Valid set: Average loss: 0.3587, Accuracy: 270/321 (84.11%)\n",
      "epoch= 430: Train set: Average loss: 0.4214, Accuracy: 1070/1283 (83.40%)\n",
      "Valid set: Average loss: 0.3512, Accuracy: 275/321 (85.67%)\n",
      "best_loss updated:  0.3512012263139089\n",
      "epoch= 440: Train set: Average loss: 0.3599, Accuracy: 1106/1283 (86.20%)\n",
      "Valid set: Average loss: 0.3147, Accuracy: 280/321 (87.23%)\n",
      "best_loss updated:  0.3147095541159312\n",
      "epoch= 450: Train set: Average loss: 0.3249, Accuracy: 1079/1283 (84.10%)\n",
      "Valid set: Average loss: 0.5865, Accuracy: 248/321 (77.26%)\n",
      "epoch= 460: Train set: Average loss: 0.3171, Accuracy: 1120/1283 (87.30%)\n",
      "Valid set: Average loss: 0.3272, Accuracy: 276/321 (85.98%)\n",
      "epoch= 470: Train set: Average loss: 0.3052, Accuracy: 1114/1283 (86.83%)\n",
      "Valid set: Average loss: 0.3514, Accuracy: 272/321 (84.74%)\n",
      "epoch= 480: Train set: Average loss: 0.4150, Accuracy: 1102/1283 (85.89%)\n",
      "Valid set: Average loss: 0.3409, Accuracy: 276/321 (85.98%)\n",
      "epoch= 490: Train set: Average loss: 0.2803, Accuracy: 1124/1283 (87.61%)\n",
      "Valid set: Average loss: 0.3391, Accuracy: 275/321 (85.67%)\n",
      "epoch= 500: Train set: Average loss: 0.3025, Accuracy: 1104/1283 (86.05%)\n",
      "Valid set: Average loss: 0.3712, Accuracy: 274/321 (85.36%)\n",
      "epoch= 510: Train set: Average loss: 0.3086, Accuracy: 1095/1283 (85.35%)\n",
      "Valid set: Average loss: 0.4251, Accuracy: 265/321 (82.55%)\n",
      "epoch= 520: Train set: Average loss: 0.2926, Accuracy: 1105/1283 (86.13%)\n",
      "Valid set: Average loss: 0.4157, Accuracy: 274/321 (85.36%)\n",
      "epoch= 530: Train set: Average loss: 0.3026, Accuracy: 1111/1283 (86.59%)\n",
      "Valid set: Average loss: 0.3090, Accuracy: 282/321 (87.85%)\n",
      "best_loss updated:  0.3089732974767685\n",
      "epoch= 540: Train set: Average loss: 0.3450, Accuracy: 1127/1283 (87.84%)\n",
      "Valid set: Average loss: 0.3032, Accuracy: 278/321 (86.60%)\n",
      "best_loss updated:  0.30318286021550495\n",
      "epoch= 550: Train set: Average loss: 0.3304, Accuracy: 1092/1283 (85.11%)\n",
      "Valid set: Average loss: 0.3656, Accuracy: 269/321 (83.80%)\n",
      "epoch= 560: Train set: Average loss: 0.3856, Accuracy: 1129/1283 (88.00%)\n",
      "Valid set: Average loss: 0.3339, Accuracy: 272/321 (84.74%)\n",
      "epoch= 570: Train set: Average loss: 0.2824, Accuracy: 1126/1283 (87.76%)\n",
      "Valid set: Average loss: 0.3312, Accuracy: 277/321 (86.29%)\n",
      "epoch= 580: Train set: Average loss: 0.3575, Accuracy: 1118/1283 (87.14%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Valid set: Average loss: 0.3218, Accuracy: 281/321 (87.54%)\n",
      "epoch= 590: Train set: Average loss: 0.2990, Accuracy: 1118/1283 (87.14%)\n",
      "Valid set: Average loss: 0.2923, Accuracy: 288/321 (89.72%)\n",
      "best_loss updated:  0.29233528176943463\n",
      "epoch= 600: Train set: Average loss: 0.2615, Accuracy: 1132/1283 (88.23%)\n",
      "Valid set: Average loss: 0.2606, Accuracy: 289/321 (90.03%)\n",
      "best_loss updated:  0.26056405901908875\n",
      "epoch= 610: Train set: Average loss: 0.2415, Accuracy: 1152/1283 (89.79%)\n",
      "Valid set: Average loss: 0.3009, Accuracy: 278/321 (86.60%)\n",
      "epoch= 620: Train set: Average loss: 0.4121, Accuracy: 1143/1283 (89.09%)\n",
      "Valid set: Average loss: 0.4263, Accuracy: 256/321 (79.75%)\n",
      "epoch= 630: Train set: Average loss: 0.2585, Accuracy: 1129/1283 (88.00%)\n",
      "Valid set: Average loss: 0.4234, Accuracy: 271/321 (84.42%)\n",
      "epoch= 640: Train set: Average loss: 0.2496, Accuracy: 1144/1283 (89.17%)\n",
      "Valid set: Average loss: 0.2878, Accuracy: 282/321 (87.85%)\n",
      "epoch= 650: Train set: Average loss: 0.2712, Accuracy: 1142/1283 (89.01%)\n",
      "Valid set: Average loss: 0.7492, Accuracy: 257/321 (80.06%)\n",
      "epoch= 660: Train set: Average loss: 0.3138, Accuracy: 1141/1283 (88.93%)\n",
      "Valid set: Average loss: 0.2975, Accuracy: 276/321 (85.98%)\n",
      "epoch= 670: Train set: Average loss: 0.2785, Accuracy: 1155/1283 (90.02%)\n",
      "Valid set: Average loss: 0.2544, Accuracy: 289/321 (90.03%)\n",
      "best_loss updated:  0.2544414500395457\n",
      "epoch= 680: Train set: Average loss: 0.2485, Accuracy: 1155/1283 (90.02%)\n",
      "Valid set: Average loss: 0.3016, Accuracy: 284/321 (88.47%)\n",
      "epoch= 690: Train set: Average loss: 0.2478, Accuracy: 1151/1283 (89.71%)\n",
      "Valid set: Average loss: 0.2759, Accuracy: 282/321 (87.85%)\n",
      "epoch= 700: Train set: Average loss: 0.3373, Accuracy: 1162/1283 (90.57%)\n",
      "Valid set: Average loss: 0.2957, Accuracy: 281/321 (87.54%)\n",
      "epoch= 710: Train set: Average loss: 0.3116, Accuracy: 1134/1283 (88.39%)\n",
      "Valid set: Average loss: 0.3899, Accuracy: 271/321 (84.42%)\n",
      "epoch= 720: Train set: Average loss: 0.3511, Accuracy: 1130/1283 (88.07%)\n",
      "Valid set: Average loss: 0.2877, Accuracy: 280/321 (87.23%)\n",
      "epoch= 730: Train set: Average loss: 0.2670, Accuracy: 1153/1283 (89.87%)\n",
      "Valid set: Average loss: 0.2720, Accuracy: 285/321 (88.79%)\n",
      "epoch= 740: Train set: Average loss: 0.2996, Accuracy: 1137/1283 (88.62%)\n",
      "Valid set: Average loss: 0.2644, Accuracy: 284/321 (88.47%)\n",
      "epoch= 750: Train set: Average loss: 0.3110, Accuracy: 1149/1283 (89.56%)\n",
      "Valid set: Average loss: 0.5914, Accuracy: 263/321 (81.93%)\n",
      "epoch= 760: Train set: Average loss: 0.2426, Accuracy: 1142/1283 (89.01%)\n",
      "Valid set: Average loss: 0.2811, Accuracy: 287/321 (89.41%)\n",
      "epoch= 770: Train set: Average loss: 0.4118, Accuracy: 1129/1283 (88.00%)\n",
      "Valid set: Average loss: 0.3962, Accuracy: 269/321 (83.80%)\n",
      "epoch= 780: Train set: Average loss: 0.2500, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.3564, Accuracy: 280/321 (87.23%)\n",
      "epoch= 790: Train set: Average loss: 0.2364, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.2722, Accuracy: 286/321 (89.10%)\n",
      "epoch= 800: Train set: Average loss: 0.2434, Accuracy: 1156/1283 (90.10%)\n",
      "Valid set: Average loss: 0.6167, Accuracy: 274/321 (85.36%)\n",
      "epoch= 810: Train set: Average loss: 0.3201, Accuracy: 1160/1283 (90.41%)\n",
      "Valid set: Average loss: 0.3043, Accuracy: 280/321 (87.23%)\n",
      "epoch= 820: Train set: Average loss: 0.3399, Accuracy: 1134/1283 (88.39%)\n",
      "Valid set: Average loss: 0.5535, Accuracy: 242/321 (75.39%)\n",
      "epoch= 830: Train set: Average loss: 0.2736, Accuracy: 1141/1283 (88.93%)\n",
      "Valid set: Average loss: 0.4002, Accuracy: 273/321 (85.05%)\n",
      "epoch= 840: Train set: Average loss: 0.2324, Accuracy: 1162/1283 (90.57%)\n",
      "Valid set: Average loss: 0.2611, Accuracy: 283/321 (88.16%)\n",
      "epoch= 850: Train set: Average loss: 0.2734, Accuracy: 1151/1283 (89.71%)\n",
      "Valid set: Average loss: 0.3855, Accuracy: 276/321 (85.98%)\n",
      "epoch= 860: Train set: Average loss: 0.2249, Accuracy: 1156/1283 (90.10%)\n",
      "Valid set: Average loss: 0.2737, Accuracy: 286/321 (89.10%)\n",
      "epoch= 870: Train set: Average loss: 0.2283, Accuracy: 1170/1283 (91.19%)\n",
      "Valid set: Average loss: 0.4656, Accuracy: 261/321 (81.31%)\n",
      "epoch= 880: Train set: Average loss: 0.2753, Accuracy: 1166/1283 (90.88%)\n",
      "Valid set: Average loss: 0.2658, Accuracy: 283/321 (88.16%)\n",
      "epoch= 890: Train set: Average loss: 0.2162, Accuracy: 1165/1283 (90.80%)\n",
      "Valid set: Average loss: 0.2624, Accuracy: 283/321 (88.16%)\n",
      "epoch= 900: Train set: Average loss: 0.2485, Accuracy: 1166/1283 (90.88%)\n",
      "Valid set: Average loss: 0.2476, Accuracy: 290/321 (90.34%)\n",
      "best_loss updated:  0.2476403017838796\n",
      "epoch= 910: Train set: Average loss: 0.2216, Accuracy: 1155/1283 (90.02%)\n",
      "Valid set: Average loss: 0.3434, Accuracy: 280/321 (87.23%)\n",
      "epoch= 920: Train set: Average loss: 0.2033, Accuracy: 1170/1283 (91.19%)\n",
      "Valid set: Average loss: 0.2762, Accuracy: 281/321 (87.54%)\n",
      "epoch= 930: Train set: Average loss: 0.2034, Accuracy: 1168/1283 (91.04%)\n",
      "Valid set: Average loss: 0.5736, Accuracy: 262/321 (81.62%)\n",
      "epoch= 940: Train set: Average loss: 0.2035, Accuracy: 1162/1283 (90.57%)\n",
      "Valid set: Average loss: 0.2847, Accuracy: 281/321 (87.54%)\n",
      "epoch= 950: Train set: Average loss: 0.3775, Accuracy: 1050/1283 (81.84%)\n",
      "Valid set: Average loss: 0.7248, Accuracy: 236/321 (73.52%)\n",
      "epoch= 960: Train set: Average loss: 0.3581, Accuracy: 1154/1283 (89.95%)\n",
      "Valid set: Average loss: 0.3130, Accuracy: 274/321 (85.36%)\n",
      "epoch= 970: Train set: Average loss: 0.2619, Accuracy: 1162/1283 (90.57%)\n",
      "Valid set: Average loss: 0.3641, Accuracy: 262/321 (81.62%)\n",
      "epoch= 980: Train set: Average loss: 0.2289, Accuracy: 1163/1283 (90.65%)\n",
      "Valid set: Average loss: 0.3046, Accuracy: 273/321 (85.05%)\n",
      "epoch= 990: Train set: Average loss: 0.3508, Accuracy: 1094/1283 (85.27%)\n",
      "Valid set: Average loss: 1.1682, Accuracy: 239/321 (74.45%)\n",
      "epoch= 1000: Train set: Average loss: 0.2175, Accuracy: 1171/1283 (91.27%)\n",
      "Valid set: Average loss: 0.2985, Accuracy: 278/321 (86.60%)\n",
      "epoch= 1010: Train set: Average loss: 0.3036, Accuracy: 1167/1283 (90.96%)\n",
      "Valid set: Average loss: 0.2710, Accuracy: 284/321 (88.47%)\n",
      "epoch= 1020: Train set: Average loss: 0.2463, Accuracy: 1151/1283 (89.71%)\n",
      "Valid set: Average loss: 0.2978, Accuracy: 278/321 (86.60%)\n",
      "epoch= 1030: Train set: Average loss: 0.2916, Accuracy: 1181/1283 (92.05%)\n",
      "Valid set: Average loss: 0.2738, Accuracy: 279/321 (86.92%)\n",
      "epoch= 1040: Train set: Average loss: 0.3145, Accuracy: 1165/1283 (90.80%)\n",
      "Valid set: Average loss: 0.2453, Accuracy: 286/321 (89.10%)\n",
      "best_loss updated:  0.24527456363042197\n",
      "epoch= 1050: Train set: Average loss: 0.2088, Accuracy: 1169/1283 (91.11%)\n",
      "Valid set: Average loss: 0.2550, Accuracy: 286/321 (89.10%)\n",
      "epoch= 1060: Train set: Average loss: 0.2054, Accuracy: 1168/1283 (91.04%)\n",
      "Valid set: Average loss: 0.2446, Accuracy: 283/321 (88.16%)\n",
      "best_loss updated:  0.2445501685142517\n",
      "epoch= 1070: Train set: Average loss: 0.2006, Accuracy: 1178/1283 (91.82%)\n",
      "Valid set: Average loss: 0.2547, Accuracy: 282/321 (87.85%)\n",
      "epoch= 1080: Train set: Average loss: 0.2608, Accuracy: 1165/1283 (90.80%)\n",
      "Valid set: Average loss: 0.2575, Accuracy: 289/321 (90.03%)\n",
      "epoch= 1090: Train set: Average loss: 0.2371, Accuracy: 1168/1283 (91.04%)\n",
      "Valid set: Average loss: 0.2396, Accuracy: 288/321 (89.72%)\n",
      "best_loss updated:  0.23955420653025308\n",
      "epoch= 1100: Train set: Average loss: 0.1961, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.2529, Accuracy: 284/321 (88.47%)\n",
      "epoch= 1110: Train set: Average loss: 0.1988, Accuracy: 1179/1283 (91.89%)\n",
      "Valid set: Average loss: 0.2520, Accuracy: 287/321 (89.41%)\n",
      "epoch= 1120: Train set: Average loss: 0.2060, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.2424, Accuracy: 286/321 (89.10%)\n",
      "epoch= 1130: Train set: Average loss: 0.2017, Accuracy: 1180/1283 (91.97%)\n",
      "Valid set: Average loss: 0.2858, Accuracy: 281/321 (87.54%)\n",
      "epoch= 1140: Train set: Average loss: 0.3113, Accuracy: 1172/1283 (91.35%)\n",
      "Valid set: Average loss: 0.2514, Accuracy: 288/321 (89.72%)\n",
      "epoch= 1150: Train set: Average loss: 0.1914, Accuracy: 1182/1283 (92.13%)\n",
      "Valid set: Average loss: 0.2584, Accuracy: 281/321 (87.54%)\n",
      "epoch= 1160: Train set: Average loss: 0.2228, Accuracy: 1177/1283 (91.74%)\n",
      "Valid set: Average loss: 0.2489, Accuracy: 283/321 (88.16%)\n",
      "epoch= 1170: Train set: Average loss: 0.2144, Accuracy: 1173/1283 (91.43%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Valid set: Average loss: 0.2538, Accuracy: 287/321 (89.41%)\n",
      "epoch= 1180: Train set: Average loss: 0.2555, Accuracy: 1177/1283 (91.74%)\n",
      "Valid set: Average loss: 0.2522, Accuracy: 284/321 (88.47%)\n",
      "epoch= 1190: Train set: Average loss: 0.2502, Accuracy: 1172/1283 (91.35%)\n",
      "Valid set: Average loss: 0.2419, Accuracy: 294/321 (91.59%)\n",
      "epoch= 1200: Train set: Average loss: 0.2071, Accuracy: 1172/1283 (91.35%)\n",
      "Valid set: Average loss: 0.2331, Accuracy: 287/321 (89.41%)\n",
      "best_loss updated:  0.23311466972033182\n",
      "epoch= 1210: Train set: Average loss: 0.2161, Accuracy: 1180/1283 (91.97%)\n",
      "Valid set: Average loss: 0.2303, Accuracy: 291/321 (90.65%)\n",
      "best_loss updated:  0.23029986023902893\n",
      "epoch= 1220: Train set: Average loss: 0.1903, Accuracy: 1176/1283 (91.66%)\n",
      "Valid set: Average loss: 0.2294, Accuracy: 291/321 (90.65%)\n",
      "best_loss updated:  0.22940454383691153\n",
      "epoch= 1230: Train set: Average loss: 0.1954, Accuracy: 1173/1283 (91.43%)\n",
      "Valid set: Average loss: 0.2347, Accuracy: 289/321 (90.03%)\n",
      "epoch= 1240: Train set: Average loss: 0.2023, Accuracy: 1183/1283 (92.21%)\n",
      "Valid set: Average loss: 0.2392, Accuracy: 290/321 (90.34%)\n",
      "epoch= 1250: Train set: Average loss: 0.2621, Accuracy: 1180/1283 (91.97%)\n",
      "Valid set: Average loss: 0.2443, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1260: Train set: Average loss: 0.1842, Accuracy: 1194/1283 (93.06%)\n",
      "Valid set: Average loss: 0.2389, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1270: Train set: Average loss: 0.2364, Accuracy: 1178/1283 (91.82%)\n",
      "Valid set: Average loss: 0.2367, Accuracy: 290/321 (90.34%)\n",
      "epoch= 1280: Train set: Average loss: 0.1870, Accuracy: 1182/1283 (92.13%)\n",
      "Valid set: Average loss: 0.2399, Accuracy: 290/321 (90.34%)\n",
      "epoch= 1290: Train set: Average loss: 0.1900, Accuracy: 1189/1283 (92.67%)\n",
      "Valid set: Average loss: 0.2431, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1300: Train set: Average loss: 0.1840, Accuracy: 1190/1283 (92.75%)\n",
      "Valid set: Average loss: 0.2383, Accuracy: 290/321 (90.34%)\n",
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EF20C7E0F0>, <torchvision.transforms.transforms.Lambda object at 0x000002EF20C7E828>, <torchvision.transforms.transforms.ToTensor object at 0x000002EF20C7EF60>]\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EF20C1FD30>, <torchvision.transforms.transforms.ToTensor object at 0x000002EF20C1F6D8>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 10: Train set: Average loss: 1.1420, Accuracy: 674/1283 (52.53%)\n",
      "Valid set: Average loss: 0.7408, Accuracy: 166/321 (51.71%)\n",
      "best_loss updated:  0.7408406535784403\n",
      "epoch= 20: Train set: Average loss: 0.7152, Accuracy: 782/1283 (60.95%)\n",
      "Valid set: Average loss: 0.6696, Accuracy: 171/321 (53.27%)\n",
      "best_loss updated:  0.6695824066797892\n",
      "epoch= 30: Train set: Average loss: 0.6486, Accuracy: 833/1283 (64.93%)\n",
      "Valid set: Average loss: 0.6172, Accuracy: 216/321 (67.29%)\n",
      "best_loss updated:  0.617192546526591\n",
      "epoch= 40: Train set: Average loss: 0.6125, Accuracy: 834/1283 (65.00%)\n",
      "Valid set: Average loss: 0.6094, Accuracy: 217/321 (67.60%)\n",
      "best_loss updated:  0.6093805432319641\n",
      "epoch= 50: Train set: Average loss: 0.6488, Accuracy: 816/1283 (63.60%)\n",
      "Valid set: Average loss: 0.6346, Accuracy: 207/321 (64.49%)\n",
      "epoch= 60: Train set: Average loss: 0.6519, Accuracy: 859/1283 (66.95%)\n",
      "Valid set: Average loss: 0.5712, Accuracy: 211/321 (65.73%)\n",
      "best_loss updated:  0.5711575150489807\n",
      "epoch= 70: Train set: Average loss: 0.5476, Accuracy: 860/1283 (67.03%)\n",
      "Valid set: Average loss: 0.5489, Accuracy: 233/321 (72.59%)\n",
      "best_loss updated:  0.5489415129025778\n",
      "epoch= 80: Train set: Average loss: 0.6368, Accuracy: 888/1283 (69.21%)\n",
      "Valid set: Average loss: 0.6213, Accuracy: 221/321 (68.85%)\n",
      "epoch= 90: Train set: Average loss: 0.5879, Accuracy: 850/1283 (66.25%)\n",
      "Valid set: Average loss: 0.6162, Accuracy: 217/321 (67.60%)\n",
      "epoch= 100: Train set: Average loss: 0.5646, Accuracy: 847/1283 (66.02%)\n",
      "Valid set: Average loss: 0.5039, Accuracy: 236/321 (73.52%)\n",
      "best_loss updated:  0.5038598477840424\n",
      "epoch= 110: Train set: Average loss: 0.6032, Accuracy: 886/1283 (69.06%)\n",
      "Valid set: Average loss: 0.6104, Accuracy: 213/321 (66.36%)\n",
      "epoch= 120: Train set: Average loss: 0.4941, Accuracy: 946/1283 (73.73%)\n",
      "Valid set: Average loss: 0.5602, Accuracy: 247/321 (76.95%)\n",
      "epoch= 130: Train set: Average loss: 0.6127, Accuracy: 891/1283 (69.45%)\n",
      "Valid set: Average loss: 0.6679, Accuracy: 218/321 (67.91%)\n",
      "epoch= 140: Train set: Average loss: 0.4687, Accuracy: 971/1283 (75.68%)\n",
      "Valid set: Average loss: 0.5373, Accuracy: 241/321 (75.08%)\n",
      "epoch= 150: Train set: Average loss: 0.4234, Accuracy: 1012/1283 (78.88%)\n",
      "Valid set: Average loss: 0.3413, Accuracy: 278/321 (86.60%)\n",
      "best_loss updated:  0.3412892321745555\n",
      "epoch= 160: Train set: Average loss: 0.5794, Accuracy: 875/1283 (68.20%)\n",
      "Valid set: Average loss: 1.3218, Accuracy: 195/321 (60.75%)\n",
      "epoch= 170: Train set: Average loss: 0.6966, Accuracy: 690/1283 (53.78%)\n",
      "Valid set: Average loss: 0.6711, Accuracy: 179/321 (55.76%)\n",
      "epoch= 180: Train set: Average loss: 0.6407, Accuracy: 811/1283 (63.21%)\n",
      "Valid set: Average loss: 0.5733, Accuracy: 221/321 (68.85%)\n",
      "epoch= 190: Train set: Average loss: 0.6245, Accuracy: 834/1283 (65.00%)\n",
      "Valid set: Average loss: 0.5556, Accuracy: 226/321 (70.40%)\n",
      "epoch= 200: Train set: Average loss: 0.5536, Accuracy: 884/1283 (68.90%)\n",
      "Valid set: Average loss: 0.5144, Accuracy: 235/321 (73.21%)\n",
      "epoch= 210: Train set: Average loss: 0.4351, Accuracy: 1006/1283 (78.41%)\n",
      "Valid set: Average loss: 0.5789, Accuracy: 238/321 (74.14%)\n",
      "epoch= 220: Train set: Average loss: 0.6312, Accuracy: 841/1283 (65.55%)\n",
      "Valid set: Average loss: 0.5357, Accuracy: 228/321 (71.03%)\n",
      "epoch= 230: Train set: Average loss: 0.4814, Accuracy: 969/1283 (75.53%)\n",
      "Valid set: Average loss: 0.4544, Accuracy: 251/321 (78.19%)\n",
      "epoch= 240: Train set: Average loss: 0.4998, Accuracy: 993/1283 (77.40%)\n",
      "Valid set: Average loss: 0.4617, Accuracy: 259/321 (80.69%)\n",
      "epoch= 250: Train set: Average loss: 0.6145, Accuracy: 832/1283 (64.85%)\n",
      "Valid set: Average loss: 0.7605, Accuracy: 172/321 (53.58%)\n",
      "epoch= 260: Train set: Average loss: 0.5860, Accuracy: 854/1283 (66.56%)\n",
      "Valid set: Average loss: 0.5612, Accuracy: 230/321 (71.65%)\n",
      "epoch= 270: Train set: Average loss: 0.4890, Accuracy: 965/1283 (75.21%)\n",
      "Valid set: Average loss: 0.5430, Accuracy: 228/321 (71.03%)\n",
      "epoch= 280: Train set: Average loss: 0.5079, Accuracy: 980/1283 (76.38%)\n",
      "Valid set: Average loss: 0.4893, Accuracy: 246/321 (76.64%)\n",
      "epoch= 290: Train set: Average loss: 0.4729, Accuracy: 997/1283 (77.71%)\n",
      "Valid set: Average loss: 0.4505, Accuracy: 260/321 (81.00%)\n",
      "epoch= 300: Train set: Average loss: 0.4548, Accuracy: 998/1283 (77.79%)\n",
      "Valid set: Average loss: 0.4153, Accuracy: 258/321 (80.37%)\n",
      "epoch= 310: Train set: Average loss: 0.4431, Accuracy: 1056/1283 (82.31%)\n",
      "Valid set: Average loss: 0.3616, Accuracy: 273/321 (85.05%)\n",
      "epoch= 320: Train set: Average loss: 0.4765, Accuracy: 1060/1283 (82.62%)\n",
      "Valid set: Average loss: 0.3386, Accuracy: 279/321 (86.92%)\n",
      "best_loss updated:  0.33855148156483966\n",
      "epoch= 330: Train set: Average loss: 0.5085, Accuracy: 1059/1283 (82.54%)\n",
      "Valid set: Average loss: 0.4156, Accuracy: 256/321 (79.75%)\n",
      "epoch= 340: Train set: Average loss: 0.4699, Accuracy: 1089/1283 (84.88%)\n",
      "Valid set: Average loss: 0.3138, Accuracy: 279/321 (86.92%)\n",
      "best_loss updated:  0.3137598733107249\n",
      "epoch= 350: Train set: Average loss: 0.3285, Accuracy: 1093/1283 (85.19%)\n",
      "Valid set: Average loss: 0.3085, Accuracy: 277/321 (86.29%)\n",
      "best_loss updated:  0.3084918061892192\n",
      "epoch= 360: Train set: Average loss: 0.3442, Accuracy: 1090/1283 (84.96%)\n",
      "Valid set: Average loss: 0.3221, Accuracy: 286/321 (89.10%)\n",
      "epoch= 370: Train set: Average loss: 0.3842, Accuracy: 1099/1283 (85.66%)\n",
      "Valid set: Average loss: 0.2761, Accuracy: 288/321 (89.72%)\n",
      "best_loss updated:  0.27607030669848126\n",
      "epoch= 380: Train set: Average loss: 0.3599, Accuracy: 1071/1283 (83.48%)\n",
      "Valid set: Average loss: 0.3054, Accuracy: 283/321 (88.16%)\n",
      "epoch= 390: Train set: Average loss: 0.2879, Accuracy: 1117/1283 (87.06%)\n",
      "Valid set: Average loss: 0.2669, Accuracy: 284/321 (88.47%)\n",
      "best_loss updated:  0.2669381449619929\n",
      "epoch= 400: Train set: Average loss: 0.3060, Accuracy: 1106/1283 (86.20%)\n",
      "Valid set: Average loss: 0.2731, Accuracy: 289/321 (90.03%)\n",
      "epoch= 410: Train set: Average loss: 0.2987, Accuracy: 1110/1283 (86.52%)\n",
      "Valid set: Average loss: 0.2545, Accuracy: 292/321 (90.97%)\n",
      "best_loss updated:  0.25453190008799237\n",
      "epoch= 420: Train set: Average loss: 0.4312, Accuracy: 1114/1283 (86.83%)\n",
      "Valid set: Average loss: 0.3384, Accuracy: 282/321 (87.85%)\n",
      "epoch= 430: Train set: Average loss: 0.3062, Accuracy: 1126/1283 (87.76%)\n",
      "Valid set: Average loss: 0.2709, Accuracy: 284/321 (88.47%)\n",
      "epoch= 440: Train set: Average loss: 0.3465, Accuracy: 1132/1283 (88.23%)\n",
      "Valid set: Average loss: 0.2729, Accuracy: 289/321 (90.03%)\n",
      "epoch= 450: Train set: Average loss: 0.4420, Accuracy: 1054/1283 (82.15%)\n",
      "Valid set: Average loss: 0.3330, Accuracy: 280/321 (87.23%)\n",
      "epoch= 460: Train set: Average loss: 0.2571, Accuracy: 1143/1283 (89.09%)\n",
      "Valid set: Average loss: 0.2690, Accuracy: 289/321 (90.03%)\n",
      "epoch= 470: Train set: Average loss: 0.3472, Accuracy: 1078/1283 (84.02%)\n",
      "Valid set: Average loss: 0.2873, Accuracy: 283/321 (88.16%)\n",
      "epoch= 480: Train set: Average loss: 0.3326, Accuracy: 1129/1283 (88.00%)\n",
      "Valid set: Average loss: 0.2899, Accuracy: 273/321 (85.05%)\n",
      "epoch= 490: Train set: Average loss: 0.3071, Accuracy: 1143/1283 (89.09%)\n",
      "Valid set: Average loss: 0.2896, Accuracy: 277/321 (86.29%)\n",
      "epoch= 500: Train set: Average loss: 0.3401, Accuracy: 1137/1283 (88.62%)\n",
      "Valid set: Average loss: 0.2613, Accuracy: 280/321 (87.23%)\n",
      "epoch= 510: Train set: Average loss: 0.2891, Accuracy: 1117/1283 (87.06%)\n",
      "Valid set: Average loss: 0.3198, Accuracy: 274/321 (85.36%)\n",
      "epoch= 520: Train set: Average loss: 0.3133, Accuracy: 1147/1283 (89.40%)\n",
      "Valid set: Average loss: 0.3042, Accuracy: 268/321 (83.49%)\n",
      "epoch= 530: Train set: Average loss: 0.3223, Accuracy: 1120/1283 (87.30%)\n",
      "Valid set: Average loss: 0.2701, Accuracy: 288/321 (89.72%)\n",
      "epoch= 540: Train set: Average loss: 0.2456, Accuracy: 1142/1283 (89.01%)\n",
      "Valid set: Average loss: 0.2460, Accuracy: 289/321 (90.03%)\n",
      "best_loss updated:  0.24598342180252075\n",
      "epoch= 550: Train set: Average loss: 0.2554, Accuracy: 1144/1283 (89.17%)\n",
      "Valid set: Average loss: 0.2340, Accuracy: 291/321 (90.65%)\n",
      "best_loss updated:  0.23398590087890625\n",
      "epoch= 560: Train set: Average loss: 0.2980, Accuracy: 1116/1283 (86.98%)\n",
      "Valid set: Average loss: 0.3448, Accuracy: 270/321 (84.11%)\n",
      "epoch= 570: Train set: Average loss: 0.2622, Accuracy: 1129/1283 (88.00%)\n",
      "Valid set: Average loss: 0.2212, Accuracy: 290/321 (90.34%)\n",
      "best_loss updated:  0.22122480471928915\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 580: Train set: Average loss: 0.4055, Accuracy: 1129/1283 (88.00%)\n",
      "Valid set: Average loss: 0.2343, Accuracy: 294/321 (91.59%)\n",
      "epoch= 590: Train set: Average loss: 0.2351, Accuracy: 1143/1283 (89.09%)\n",
      "Valid set: Average loss: 0.2329, Accuracy: 284/321 (88.47%)\n",
      "epoch= 600: Train set: Average loss: 0.2308, Accuracy: 1178/1283 (91.82%)\n",
      "Valid set: Average loss: 0.2174, Accuracy: 286/321 (89.10%)\n",
      "best_loss updated:  0.21739680071671805\n",
      "epoch= 610: Train set: Average loss: 0.2102, Accuracy: 1170/1283 (91.19%)\n",
      "Valid set: Average loss: 0.2499, Accuracy: 285/321 (88.79%)\n",
      "epoch= 620: Train set: Average loss: 0.3029, Accuracy: 1122/1283 (87.45%)\n",
      "Valid set: Average loss: 0.4339, Accuracy: 260/321 (81.00%)\n",
      "epoch= 630: Train set: Average loss: 0.3365, Accuracy: 1160/1283 (90.41%)\n",
      "Valid set: Average loss: 0.2646, Accuracy: 281/321 (87.54%)\n",
      "epoch= 640: Train set: Average loss: 0.3526, Accuracy: 1145/1283 (89.24%)\n",
      "Valid set: Average loss: 0.2411, Accuracy: 291/321 (90.65%)\n",
      "epoch= 650: Train set: Average loss: 0.3080, Accuracy: 1159/1283 (90.34%)\n",
      "Valid set: Average loss: 0.2422, Accuracy: 294/321 (91.59%)\n",
      "epoch= 660: Train set: Average loss: 0.2043, Accuracy: 1176/1283 (91.66%)\n",
      "Valid set: Average loss: 0.2406, Accuracy: 288/321 (89.72%)\n",
      "epoch= 670: Train set: Average loss: 0.3350, Accuracy: 1092/1283 (85.11%)\n",
      "Valid set: Average loss: 0.3101, Accuracy: 276/321 (85.98%)\n",
      "epoch= 680: Train set: Average loss: 0.2188, Accuracy: 1156/1283 (90.10%)\n",
      "Valid set: Average loss: 0.2219, Accuracy: 290/321 (90.34%)\n",
      "epoch= 690: Train set: Average loss: 0.2097, Accuracy: 1177/1283 (91.74%)\n",
      "Valid set: Average loss: 0.2716, Accuracy: 290/321 (90.34%)\n",
      "epoch= 700: Train set: Average loss: 0.1907, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.2778, Accuracy: 282/321 (87.85%)\n",
      "epoch= 710: Train set: Average loss: 0.4683, Accuracy: 977/1283 (76.15%)\n",
      "Valid set: Average loss: 0.4496, Accuracy: 250/321 (77.88%)\n",
      "epoch= 720: Train set: Average loss: 0.2721, Accuracy: 1123/1283 (87.53%)\n",
      "Valid set: Average loss: 0.2618, Accuracy: 282/321 (87.85%)\n",
      "epoch= 730: Train set: Average loss: 0.4352, Accuracy: 1112/1283 (86.67%)\n",
      "Valid set: Average loss: 0.2586, Accuracy: 291/321 (90.65%)\n",
      "epoch= 740: Train set: Average loss: 0.3142, Accuracy: 1135/1283 (88.46%)\n",
      "Valid set: Average loss: 0.2909, Accuracy: 270/321 (84.11%)\n",
      "epoch= 750: Train set: Average loss: 0.3691, Accuracy: 1132/1283 (88.23%)\n",
      "Valid set: Average loss: 0.2820, Accuracy: 274/321 (85.36%)\n",
      "epoch= 760: Train set: Average loss: 0.3183, Accuracy: 1165/1283 (90.80%)\n",
      "Valid set: Average loss: 0.2302, Accuracy: 293/321 (91.28%)\n",
      "epoch= 770: Train set: Average loss: 0.2269, Accuracy: 1157/1283 (90.18%)\n",
      "Valid set: Average loss: 0.3899, Accuracy: 257/321 (80.06%)\n",
      "epoch= 780: Train set: Average loss: 0.2582, Accuracy: 1130/1283 (88.07%)\n",
      "Valid set: Average loss: 0.2886, Accuracy: 283/321 (88.16%)\n",
      "epoch= 790: Train set: Average loss: 0.2508, Accuracy: 1155/1283 (90.02%)\n",
      "Valid set: Average loss: 0.2652, Accuracy: 285/321 (88.79%)\n",
      "epoch= 800: Train set: Average loss: 0.2366, Accuracy: 1152/1283 (89.79%)\n",
      "Valid set: Average loss: 0.2653, Accuracy: 283/321 (88.16%)\n",
      "epoch= 810: Train set: Average loss: 0.2197, Accuracy: 1163/1283 (90.65%)\n",
      "Valid set: Average loss: 0.2039, Accuracy: 292/321 (90.97%)\n",
      "best_loss updated:  0.20393107334772745\n",
      "epoch= 820: Train set: Average loss: 0.3409, Accuracy: 1163/1283 (90.65%)\n",
      "Valid set: Average loss: 0.2457, Accuracy: 285/321 (88.79%)\n",
      "epoch= 830: Train set: Average loss: 0.3304, Accuracy: 1122/1283 (87.45%)\n",
      "Valid set: Average loss: 1.1623, Accuracy: 180/321 (56.07%)\n",
      "epoch= 840: Train set: Average loss: 0.2237, Accuracy: 1166/1283 (90.88%)\n",
      "Valid set: Average loss: 0.2177, Accuracy: 295/321 (91.90%)\n",
      "epoch= 850: Train set: Average loss: 0.2202, Accuracy: 1157/1283 (90.18%)\n",
      "Valid set: Average loss: 0.2949, Accuracy: 279/321 (86.92%)\n",
      "epoch= 860: Train set: Average loss: 0.2644, Accuracy: 1175/1283 (91.58%)\n",
      "Valid set: Average loss: 0.2073, Accuracy: 298/321 (92.83%)\n",
      "epoch= 870: Train set: Average loss: 0.2525, Accuracy: 1129/1283 (88.00%)\n",
      "Valid set: Average loss: 0.2520, Accuracy: 284/321 (88.47%)\n",
      "epoch= 880: Train set: Average loss: 0.3551, Accuracy: 1170/1283 (91.19%)\n",
      "Valid set: Average loss: 0.2603, Accuracy: 287/321 (89.41%)\n",
      "epoch= 890: Train set: Average loss: 0.2056, Accuracy: 1181/1283 (92.05%)\n",
      "Valid set: Average loss: 0.2282, Accuracy: 289/321 (90.03%)\n",
      "epoch= 900: Train set: Average loss: 0.2194, Accuracy: 1178/1283 (91.82%)\n",
      "Valid set: Average loss: 0.2659, Accuracy: 286/321 (89.10%)\n",
      "epoch= 910: Train set: Average loss: 0.2167, Accuracy: 1176/1283 (91.66%)\n",
      "Valid set: Average loss: 0.3664, Accuracy: 265/321 (82.55%)\n",
      "epoch= 920: Train set: Average loss: 0.2222, Accuracy: 1164/1283 (90.72%)\n",
      "Valid set: Average loss: 0.2724, Accuracy: 289/321 (90.03%)\n",
      "epoch= 930: Train set: Average loss: 0.1804, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.2281, Accuracy: 286/321 (89.10%)\n",
      "epoch= 940: Train set: Average loss: 0.1733, Accuracy: 1184/1283 (92.28%)\n",
      "Valid set: Average loss: 0.3029, Accuracy: 286/321 (89.10%)\n",
      "epoch= 950: Train set: Average loss: 0.2841, Accuracy: 1161/1283 (90.49%)\n",
      "Valid set: Average loss: 0.4342, Accuracy: 260/321 (81.00%)\n",
      "epoch= 960: Train set: Average loss: 0.2633, Accuracy: 1173/1283 (91.43%)\n",
      "Valid set: Average loss: 0.3823, Accuracy: 272/321 (84.74%)\n",
      "epoch= 970: Train set: Average loss: 0.1827, Accuracy: 1194/1283 (93.06%)\n",
      "Valid set: Average loss: 0.1984, Accuracy: 295/321 (91.90%)\n",
      "best_loss updated:  0.19840307533740997\n",
      "epoch= 980: Train set: Average loss: 0.2339, Accuracy: 1168/1283 (91.04%)\n",
      "Valid set: Average loss: 0.4770, Accuracy: 248/321 (77.26%)\n",
      "epoch= 990: Train set: Average loss: 0.2121, Accuracy: 1168/1283 (91.04%)\n",
      "Valid set: Average loss: 0.5063, Accuracy: 254/321 (79.13%)\n",
      "epoch= 1000: Train set: Average loss: 0.1752, Accuracy: 1193/1283 (92.99%)\n",
      "Valid set: Average loss: 0.2308, Accuracy: 293/321 (91.28%)\n",
      "epoch= 1010: Train set: Average loss: 0.1923, Accuracy: 1179/1283 (91.89%)\n",
      "Valid set: Average loss: 0.2199, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1020: Train set: Average loss: 0.1968, Accuracy: 1180/1283 (91.97%)\n",
      "Valid set: Average loss: 0.2444, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1030: Train set: Average loss: 0.1663, Accuracy: 1208/1283 (94.15%)\n",
      "Valid set: Average loss: 0.2258, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1040: Train set: Average loss: 0.1775, Accuracy: 1194/1283 (93.06%)\n",
      "Valid set: Average loss: 0.2248, Accuracy: 294/321 (91.59%)\n",
      "epoch= 1050: Train set: Average loss: 0.3019, Accuracy: 1195/1283 (93.14%)\n",
      "Valid set: Average loss: 0.2143, Accuracy: 294/321 (91.59%)\n",
      "epoch= 1060: Train set: Average loss: 0.1698, Accuracy: 1199/1283 (93.45%)\n",
      "Valid set: Average loss: 0.2041, Accuracy: 297/321 (92.52%)\n",
      "epoch= 1070: Train set: Average loss: 0.2037, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.2076, Accuracy: 296/321 (92.21%)\n",
      "epoch= 1080: Train set: Average loss: 0.2019, Accuracy: 1202/1283 (93.69%)\n",
      "Valid set: Average loss: 0.2204, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1090: Train set: Average loss: 0.2193, Accuracy: 1189/1283 (92.67%)\n",
      "Valid set: Average loss: 0.2594, Accuracy: 286/321 (89.10%)\n",
      "epoch= 1100: Train set: Average loss: 0.2169, Accuracy: 1199/1283 (93.45%)\n",
      "Valid set: Average loss: 0.2151, Accuracy: 296/321 (92.21%)\n",
      "epoch= 1110: Train set: Average loss: 0.2793, Accuracy: 1200/1283 (93.53%)\n",
      "Valid set: Average loss: 0.2594, Accuracy: 289/321 (90.03%)\n",
      "epoch= 1120: Train set: Average loss: 0.2555, Accuracy: 1188/1283 (92.60%)\n",
      "Valid set: Average loss: 0.2252, Accuracy: 294/321 (91.59%)\n",
      "epoch= 1130: Train set: Average loss: 0.1715, Accuracy: 1206/1283 (94.00%)\n",
      "Valid set: Average loss: 0.2702, Accuracy: 286/321 (89.10%)\n",
      "epoch= 1140: Train set: Average loss: 0.1651, Accuracy: 1195/1283 (93.14%)\n",
      "Valid set: Average loss: 0.2221, Accuracy: 295/321 (91.90%)\n",
      "epoch= 1150: Train set: Average loss: 0.2214, Accuracy: 1206/1283 (94.00%)\n",
      "Valid set: Average loss: 0.2358, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1160: Train set: Average loss: 0.2297, Accuracy: 1208/1283 (94.15%)\n",
      "Valid set: Average loss: 0.2091, Accuracy: 297/321 (92.52%)\n",
      "epoch= 1170: Train set: Average loss: 0.2171, Accuracy: 1192/1283 (92.91%)\n",
      "Valid set: Average loss: 0.1989, Accuracy: 299/321 (93.15%)\n",
      "epoch= 1180: Train set: Average loss: 0.1550, Accuracy: 1200/1283 (93.53%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Valid set: Average loss: 0.2152, Accuracy: 298/321 (92.83%)\n",
      "epoch= 1190: Train set: Average loss: 0.1790, Accuracy: 1214/1283 (94.62%)\n",
      "Valid set: Average loss: 0.2392, Accuracy: 289/321 (90.03%)\n",
      "epoch= 1200: Train set: Average loss: 0.1698, Accuracy: 1201/1283 (93.61%)\n",
      "Valid set: Average loss: 0.2512, Accuracy: 285/321 (88.79%)\n",
      "epoch= 1210: Train set: Average loss: 0.1615, Accuracy: 1199/1283 (93.45%)\n",
      "Valid set: Average loss: 0.2101, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1220: Train set: Average loss: 0.1580, Accuracy: 1197/1283 (93.30%)\n",
      "Valid set: Average loss: 0.2295, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1230: Train set: Average loss: 0.2495, Accuracy: 1201/1283 (93.61%)\n",
      "Valid set: Average loss: 0.2250, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1240: Train set: Average loss: 0.1542, Accuracy: 1197/1283 (93.30%)\n",
      "Valid set: Average loss: 0.2118, Accuracy: 294/321 (91.59%)\n",
      "epoch= 1250: Train set: Average loss: 0.2023, Accuracy: 1213/1283 (94.54%)\n",
      "Valid set: Average loss: 0.2214, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1260: Train set: Average loss: 0.1420, Accuracy: 1219/1283 (95.01%)\n",
      "Valid set: Average loss: 0.2177, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1270: Train set: Average loss: 0.1482, Accuracy: 1207/1283 (94.08%)\n",
      "Valid set: Average loss: 0.2098, Accuracy: 294/321 (91.59%)\n",
      "epoch= 1280: Train set: Average loss: 0.1609, Accuracy: 1200/1283 (93.53%)\n",
      "Valid set: Average loss: 0.2042, Accuracy: 299/321 (93.15%)\n",
      "epoch= 1290: Train set: Average loss: 0.2749, Accuracy: 1208/1283 (94.15%)\n",
      "Valid set: Average loss: 0.2328, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1300: Train set: Average loss: 0.2409, Accuracy: 1206/1283 (94.00%)\n",
      "Valid set: Average loss: 0.2144, Accuracy: 295/321 (91.90%)\n",
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EF41E71198>, <torchvision.transforms.transforms.Lambda object at 0x000002EF41E712B0>, <torchvision.transforms.transforms.ToTensor object at 0x000002EF41E71668>]\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EF41E71860>, <torchvision.transforms.transforms.ToTensor object at 0x000002EF41E71898>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 10: Train set: Average loss: 0.7263, Accuracy: 712/1283 (55.49%)\n",
      "Valid set: Average loss: 0.8693, Accuracy: 170/321 (52.96%)\n",
      "best_loss updated:  0.8693363269170126\n",
      "epoch= 20: Train set: Average loss: 0.5457, Accuracy: 845/1283 (65.86%)\n",
      "Valid set: Average loss: 0.6889, Accuracy: 172/321 (53.58%)\n",
      "best_loss updated:  0.6888925035794576\n",
      "epoch= 30: Train set: Average loss: 0.5984, Accuracy: 873/1283 (68.04%)\n",
      "Valid set: Average loss: 0.6013, Accuracy: 197/321 (61.37%)\n",
      "best_loss updated:  0.6013276775677999\n",
      "epoch= 40: Train set: Average loss: 0.5319, Accuracy: 879/1283 (68.51%)\n",
      "Valid set: Average loss: 0.6918, Accuracy: 182/321 (56.70%)\n",
      "epoch= 50: Train set: Average loss: 0.4847, Accuracy: 994/1283 (77.47%)\n",
      "Valid set: Average loss: 0.8246, Accuracy: 152/321 (47.35%)\n",
      "epoch= 60: Train set: Average loss: 0.4889, Accuracy: 961/1283 (74.90%)\n",
      "Valid set: Average loss: 0.4650, Accuracy: 252/321 (78.50%)\n",
      "best_loss updated:  0.46499759952227276\n",
      "epoch= 70: Train set: Average loss: 0.5597, Accuracy: 1003/1283 (78.18%)\n",
      "Valid set: Average loss: 0.6746, Accuracy: 174/321 (54.21%)\n",
      "epoch= 80: Train set: Average loss: 0.5068, Accuracy: 955/1283 (74.43%)\n",
      "Valid set: Average loss: 0.8704, Accuracy: 206/321 (64.17%)\n",
      "epoch= 90: Train set: Average loss: 0.4565, Accuracy: 1015/1283 (79.11%)\n",
      "Valid set: Average loss: 0.5236, Accuracy: 256/321 (79.75%)\n",
      "epoch= 100: Train set: Average loss: 0.4779, Accuracy: 977/1283 (76.15%)\n",
      "Valid set: Average loss: 0.5596, Accuracy: 240/321 (74.77%)\n",
      "epoch= 110: Train set: Average loss: 0.4782, Accuracy: 971/1283 (75.68%)\n",
      "Valid set: Average loss: 0.6289, Accuracy: 216/321 (67.29%)\n",
      "epoch= 120: Train set: Average loss: 0.5262, Accuracy: 1002/1283 (78.10%)\n",
      "Valid set: Average loss: 0.4683, Accuracy: 263/321 (81.93%)\n",
      "epoch= 130: Train set: Average loss: 0.5101, Accuracy: 990/1283 (77.16%)\n",
      "Valid set: Average loss: 0.4898, Accuracy: 244/321 (76.01%)\n",
      "epoch= 140: Train set: Average loss: 0.4282, Accuracy: 1012/1283 (78.88%)\n",
      "Valid set: Average loss: 0.4978, Accuracy: 253/321 (78.82%)\n",
      "epoch= 150: Train set: Average loss: 0.4484, Accuracy: 1009/1283 (78.64%)\n",
      "Valid set: Average loss: 0.4320, Accuracy: 263/321 (81.93%)\n",
      "best_loss updated:  0.431956946849823\n",
      "epoch= 160: Train set: Average loss: 0.4688, Accuracy: 967/1283 (75.37%)\n",
      "Valid set: Average loss: 0.4406, Accuracy: 258/321 (80.37%)\n",
      "epoch= 170: Train set: Average loss: 0.4484, Accuracy: 982/1283 (76.54%)\n",
      "Valid set: Average loss: 0.4638, Accuracy: 252/321 (78.50%)\n",
      "epoch= 180: Train set: Average loss: 0.4489, Accuracy: 1006/1283 (78.41%)\n",
      "Valid set: Average loss: 0.4830, Accuracy: 248/321 (77.26%)\n",
      "epoch= 190: Train set: Average loss: 0.4425, Accuracy: 1005/1283 (78.33%)\n",
      "Valid set: Average loss: 0.4114, Accuracy: 263/321 (81.93%)\n",
      "best_loss updated:  0.41144781311353046\n",
      "epoch= 200: Train set: Average loss: 0.4129, Accuracy: 1066/1283 (83.09%)\n",
      "Valid set: Average loss: 0.3451, Accuracy: 279/321 (86.92%)\n",
      "best_loss updated:  0.3450605670611064\n",
      "epoch= 210: Train set: Average loss: 0.4814, Accuracy: 1007/1283 (78.49%)\n",
      "Valid set: Average loss: 2.6500, Accuracy: 175/321 (54.52%)\n",
      "epoch= 220: Train set: Average loss: 0.3906, Accuracy: 1068/1283 (83.24%)\n",
      "Valid set: Average loss: 0.3438, Accuracy: 278/321 (86.60%)\n",
      "best_loss updated:  0.34381067752838135\n",
      "epoch= 230: Train set: Average loss: 0.3975, Accuracy: 1053/1283 (82.07%)\n",
      "Valid set: Average loss: 0.3961, Accuracy: 269/321 (83.80%)\n",
      "epoch= 240: Train set: Average loss: 0.4576, Accuracy: 1063/1283 (82.85%)\n",
      "Valid set: Average loss: 0.4515, Accuracy: 251/321 (78.19%)\n",
      "epoch= 250: Train set: Average loss: 0.4182, Accuracy: 1069/1283 (83.32%)\n",
      "Valid set: Average loss: 0.3904, Accuracy: 275/321 (85.67%)\n",
      "epoch= 260: Train set: Average loss: 0.4507, Accuracy: 1012/1283 (78.88%)\n",
      "Valid set: Average loss: 0.4711, Accuracy: 240/321 (74.77%)\n",
      "epoch= 270: Train set: Average loss: 0.3859, Accuracy: 1056/1283 (82.31%)\n",
      "Valid set: Average loss: 0.3668, Accuracy: 272/321 (84.74%)\n",
      "epoch= 280: Train set: Average loss: 0.4119, Accuracy: 1030/1283 (80.28%)\n",
      "Valid set: Average loss: 0.4017, Accuracy: 264/321 (82.24%)\n",
      "epoch= 290: Train set: Average loss: 0.4121, Accuracy: 1050/1283 (81.84%)\n",
      "Valid set: Average loss: 0.4287, Accuracy: 256/321 (79.75%)\n",
      "epoch= 300: Train set: Average loss: 0.4690, Accuracy: 1072/1283 (83.55%)\n",
      "Valid set: Average loss: 0.3538, Accuracy: 274/321 (85.36%)\n",
      "epoch= 310: Train set: Average loss: 0.3815, Accuracy: 1071/1283 (83.48%)\n",
      "Valid set: Average loss: 0.3521, Accuracy: 273/321 (85.05%)\n",
      "epoch= 320: Train set: Average loss: 0.4504, Accuracy: 1049/1283 (81.76%)\n",
      "Valid set: Average loss: 0.3408, Accuracy: 279/321 (86.92%)\n",
      "best_loss updated:  0.34084904193878174\n",
      "epoch= 330: Train set: Average loss: 0.3552, Accuracy: 1078/1283 (84.02%)\n",
      "Valid set: Average loss: 0.3930, Accuracy: 270/321 (84.11%)\n",
      "epoch= 340: Train set: Average loss: 0.3187, Accuracy: 1103/1283 (85.97%)\n",
      "Valid set: Average loss: 0.3417, Accuracy: 277/321 (86.29%)\n",
      "epoch= 350: Train set: Average loss: 0.3910, Accuracy: 1115/1283 (86.91%)\n",
      "Valid set: Average loss: 0.3303, Accuracy: 279/321 (86.92%)\n",
      "best_loss updated:  0.3302984833717346\n",
      "epoch= 360: Train set: Average loss: 0.3201, Accuracy: 1100/1283 (85.74%)\n",
      "Valid set: Average loss: 0.3293, Accuracy: 278/321 (86.60%)\n",
      "best_loss updated:  0.3293394247690837\n",
      "epoch= 370: Train set: Average loss: 0.3708, Accuracy: 1102/1283 (85.89%)\n",
      "Valid set: Average loss: 0.3615, Accuracy: 268/321 (83.49%)\n",
      "epoch= 380: Train set: Average loss: 0.3107, Accuracy: 1111/1283 (86.59%)\n",
      "Valid set: Average loss: 0.3426, Accuracy: 281/321 (87.54%)\n",
      "epoch= 390: Train set: Average loss: 0.3093, Accuracy: 1113/1283 (86.75%)\n",
      "Valid set: Average loss: 0.3239, Accuracy: 277/321 (86.29%)\n",
      "best_loss updated:  0.3239029347896576\n",
      "epoch= 400: Train set: Average loss: 0.3155, Accuracy: 1114/1283 (86.83%)\n",
      "Valid set: Average loss: 0.3231, Accuracy: 281/321 (87.54%)\n",
      "best_loss updated:  0.323069562514623\n",
      "epoch= 410: Train set: Average loss: 0.4243, Accuracy: 1116/1283 (86.98%)\n",
      "Valid set: Average loss: 0.3191, Accuracy: 277/321 (86.29%)\n",
      "best_loss updated:  0.31908130645751953\n",
      "epoch= 420: Train set: Average loss: 0.3743, Accuracy: 1114/1283 (86.83%)\n",
      "Valid set: Average loss: 0.3198, Accuracy: 273/321 (85.05%)\n",
      "epoch= 430: Train set: Average loss: 0.3772, Accuracy: 1111/1283 (86.59%)\n",
      "Valid set: Average loss: 0.3211, Accuracy: 276/321 (85.98%)\n",
      "epoch= 440: Train set: Average loss: 0.3382, Accuracy: 1109/1283 (86.44%)\n",
      "Valid set: Average loss: 0.3099, Accuracy: 276/321 (85.98%)\n",
      "best_loss updated:  0.3098847468694051\n",
      "epoch= 450: Train set: Average loss: 0.3285, Accuracy: 1124/1283 (87.61%)\n",
      "Valid set: Average loss: 0.3091, Accuracy: 282/321 (87.85%)\n",
      "best_loss updated:  0.30914240578810376\n",
      "epoch= 460: Train set: Average loss: 0.3413, Accuracy: 1107/1283 (86.28%)\n",
      "Valid set: Average loss: 0.2978, Accuracy: 279/321 (86.92%)\n",
      "best_loss updated:  0.29777393738428753\n",
      "epoch= 470: Train set: Average loss: 0.3166, Accuracy: 1127/1283 (87.84%)\n",
      "Valid set: Average loss: 0.3682, Accuracy: 277/321 (86.29%)\n",
      "epoch= 480: Train set: Average loss: 0.3256, Accuracy: 1118/1283 (87.14%)\n",
      "Valid set: Average loss: 0.3164, Accuracy: 280/321 (87.23%)\n",
      "epoch= 490: Train set: Average loss: 0.3045, Accuracy: 1107/1283 (86.28%)\n",
      "Valid set: Average loss: 0.3247, Accuracy: 279/321 (86.92%)\n",
      "epoch= 500: Train set: Average loss: 0.3248, Accuracy: 1095/1283 (85.35%)\n",
      "Valid set: Average loss: 0.3670, Accuracy: 282/321 (87.85%)\n",
      "epoch= 510: Train set: Average loss: 0.2781, Accuracy: 1116/1283 (86.98%)\n",
      "Valid set: Average loss: 0.3029, Accuracy: 281/321 (87.54%)\n",
      "epoch= 520: Train set: Average loss: 0.2556, Accuracy: 1133/1283 (88.31%)\n",
      "Valid set: Average loss: 0.2721, Accuracy: 283/321 (88.16%)\n",
      "best_loss updated:  0.27210383613904315\n",
      "epoch= 530: Train set: Average loss: 0.2876, Accuracy: 1146/1283 (89.32%)\n",
      "Valid set: Average loss: 0.3012, Accuracy: 284/321 (88.47%)\n",
      "epoch= 540: Train set: Average loss: 0.2695, Accuracy: 1132/1283 (88.23%)\n",
      "Valid set: Average loss: 0.2707, Accuracy: 280/321 (87.23%)\n",
      "best_loss updated:  0.27072663108507794\n",
      "epoch= 550: Train set: Average loss: 0.2827, Accuracy: 1126/1283 (87.76%)\n",
      "Valid set: Average loss: 0.2638, Accuracy: 283/321 (88.16%)\n",
      "best_loss updated:  0.2638120303551356\n",
      "epoch= 560: Train set: Average loss: 0.2869, Accuracy: 1137/1283 (88.62%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Valid set: Average loss: 0.2720, Accuracy: 279/321 (86.92%)\n",
      "epoch= 570: Train set: Average loss: 0.3292, Accuracy: 1161/1283 (90.49%)\n",
      "Valid set: Average loss: 0.2753, Accuracy: 280/321 (87.23%)\n",
      "epoch= 580: Train set: Average loss: 0.3024, Accuracy: 1145/1283 (89.24%)\n",
      "Valid set: Average loss: 0.4847, Accuracy: 266/321 (82.87%)\n",
      "epoch= 590: Train set: Average loss: 0.2344, Accuracy: 1157/1283 (90.18%)\n",
      "Valid set: Average loss: 0.2707, Accuracy: 282/321 (87.85%)\n",
      "epoch= 600: Train set: Average loss: 0.2396, Accuracy: 1159/1283 (90.34%)\n",
      "Valid set: Average loss: 0.2792, Accuracy: 288/321 (89.72%)\n",
      "epoch= 610: Train set: Average loss: 0.2448, Accuracy: 1150/1283 (89.63%)\n",
      "Valid set: Average loss: 0.3204, Accuracy: 281/321 (87.54%)\n",
      "epoch= 620: Train set: Average loss: 0.2400, Accuracy: 1151/1283 (89.71%)\n",
      "Valid set: Average loss: 0.2601, Accuracy: 287/321 (89.41%)\n",
      "best_loss updated:  0.2600688288609187\n",
      "epoch= 630: Train set: Average loss: 0.2524, Accuracy: 1145/1283 (89.24%)\n",
      "Valid set: Average loss: 0.2919, Accuracy: 284/321 (88.47%)\n",
      "epoch= 640: Train set: Average loss: 0.2808, Accuracy: 1129/1283 (88.00%)\n",
      "Valid set: Average loss: 0.2654, Accuracy: 284/321 (88.47%)\n",
      "epoch= 650: Train set: Average loss: 0.2331, Accuracy: 1153/1283 (89.87%)\n",
      "Valid set: Average loss: 0.2452, Accuracy: 289/321 (90.03%)\n",
      "best_loss updated:  0.24521283308664957\n",
      "epoch= 660: Train set: Average loss: 0.3467, Accuracy: 1139/1283 (88.78%)\n",
      "Valid set: Average loss: 0.2904, Accuracy: 278/321 (86.60%)\n",
      "epoch= 670: Train set: Average loss: 0.2750, Accuracy: 1147/1283 (89.40%)\n",
      "Valid set: Average loss: 0.3434, Accuracy: 280/321 (87.23%)\n",
      "epoch= 680: Train set: Average loss: 0.3976, Accuracy: 1115/1283 (86.91%)\n",
      "Valid set: Average loss: 0.3410, Accuracy: 272/321 (84.74%)\n",
      "epoch= 690: Train set: Average loss: 0.2523, Accuracy: 1150/1283 (89.63%)\n",
      "Valid set: Average loss: 0.2722, Accuracy: 282/321 (87.85%)\n",
      "epoch= 700: Train set: Average loss: 0.2854, Accuracy: 1149/1283 (89.56%)\n",
      "Valid set: Average loss: 1.5667, Accuracy: 175/321 (54.52%)\n",
      "epoch= 710: Train set: Average loss: 0.2482, Accuracy: 1139/1283 (88.78%)\n",
      "Valid set: Average loss: 0.2797, Accuracy: 276/321 (85.98%)\n",
      "epoch= 720: Train set: Average loss: 0.2230, Accuracy: 1171/1283 (91.27%)\n",
      "Valid set: Average loss: 0.2831, Accuracy: 279/321 (86.92%)\n",
      "epoch= 730: Train set: Average loss: 0.2183, Accuracy: 1166/1283 (90.88%)\n",
      "Valid set: Average loss: 0.2459, Accuracy: 287/321 (89.41%)\n",
      "epoch= 740: Train set: Average loss: 0.2592, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.2613, Accuracy: 285/321 (88.79%)\n",
      "epoch= 750: Train set: Average loss: 0.3330, Accuracy: 1163/1283 (90.65%)\n",
      "Valid set: Average loss: 0.2748, Accuracy: 288/321 (89.72%)\n",
      "epoch= 760: Train set: Average loss: 0.2757, Accuracy: 1148/1283 (89.48%)\n",
      "Valid set: Average loss: 0.2888, Accuracy: 285/321 (88.79%)\n",
      "epoch= 770: Train set: Average loss: 0.2799, Accuracy: 1159/1283 (90.34%)\n",
      "Valid set: Average loss: 0.9527, Accuracy: 201/321 (62.62%)\n",
      "epoch= 780: Train set: Average loss: 0.2763, Accuracy: 1135/1283 (88.46%)\n",
      "Valid set: Average loss: 0.3090, Accuracy: 275/321 (85.67%)\n",
      "epoch= 790: Train set: Average loss: 0.2101, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.2457, Accuracy: 289/321 (90.03%)\n",
      "epoch= 800: Train set: Average loss: 0.3136, Accuracy: 1149/1283 (89.56%)\n",
      "Valid set: Average loss: 0.3876, Accuracy: 278/321 (86.60%)\n",
      "epoch= 810: Train set: Average loss: 0.2426, Accuracy: 1156/1283 (90.10%)\n",
      "Valid set: Average loss: 0.2505, Accuracy: 286/321 (89.10%)\n",
      "epoch= 820: Train set: Average loss: 0.3556, Accuracy: 1149/1283 (89.56%)\n",
      "Valid set: Average loss: 0.2861, Accuracy: 285/321 (88.79%)\n",
      "epoch= 830: Train set: Average loss: 0.2762, Accuracy: 1153/1283 (89.87%)\n",
      "Valid set: Average loss: 0.2477, Accuracy: 287/321 (89.41%)\n",
      "epoch= 840: Train set: Average loss: 0.2280, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.2575, Accuracy: 284/321 (88.47%)\n",
      "epoch= 850: Train set: Average loss: 0.2452, Accuracy: 1173/1283 (91.43%)\n",
      "Valid set: Average loss: 0.2608, Accuracy: 283/321 (88.16%)\n",
      "epoch= 860: Train set: Average loss: 0.3716, Accuracy: 1146/1283 (89.32%)\n",
      "Valid set: Average loss: 0.2965, Accuracy: 283/321 (88.16%)\n",
      "epoch= 870: Train set: Average loss: 0.2770, Accuracy: 1151/1283 (89.71%)\n",
      "Valid set: Average loss: 0.2356, Accuracy: 289/321 (90.03%)\n",
      "best_loss updated:  0.23563143610954285\n",
      "epoch= 880: Train set: Average loss: 0.2475, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.2869, Accuracy: 285/321 (88.79%)\n",
      "epoch= 890: Train set: Average loss: 0.2841, Accuracy: 1168/1283 (91.04%)\n",
      "Valid set: Average loss: 0.2606, Accuracy: 286/321 (89.10%)\n",
      "epoch= 900: Train set: Average loss: 0.2409, Accuracy: 1151/1283 (89.71%)\n",
      "Valid set: Average loss: 0.2409, Accuracy: 286/321 (89.10%)\n",
      "epoch= 910: Train set: Average loss: 0.2135, Accuracy: 1167/1283 (90.96%)\n",
      "Valid set: Average loss: 0.2560, Accuracy: 288/321 (89.72%)\n",
      "epoch= 920: Train set: Average loss: 0.2020, Accuracy: 1179/1283 (91.89%)\n",
      "Valid set: Average loss: 0.2256, Accuracy: 289/321 (90.03%)\n",
      "best_loss updated:  0.2255520522594452\n",
      "epoch= 930: Train set: Average loss: 0.2845, Accuracy: 1125/1283 (87.69%)\n",
      "Valid set: Average loss: 0.3015, Accuracy: 280/321 (87.23%)\n",
      "epoch= 940: Train set: Average loss: 0.2260, Accuracy: 1171/1283 (91.27%)\n",
      "Valid set: Average loss: 0.2543, Accuracy: 295/321 (91.90%)\n",
      "epoch= 950: Train set: Average loss: 0.3382, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.2525, Accuracy: 284/321 (88.47%)\n",
      "epoch= 960: Train set: Average loss: 0.2614, Accuracy: 1170/1283 (91.19%)\n",
      "Valid set: Average loss: 0.3616, Accuracy: 275/321 (85.67%)\n",
      "epoch= 970: Train set: Average loss: 0.2487, Accuracy: 1153/1283 (89.87%)\n",
      "Valid set: Average loss: 0.2598, Accuracy: 281/321 (87.54%)\n",
      "epoch= 980: Train set: Average loss: 0.2935, Accuracy: 1150/1283 (89.63%)\n",
      "Valid set: Average loss: 0.2273, Accuracy: 290/321 (90.34%)\n",
      "epoch= 990: Train set: Average loss: 0.2301, Accuracy: 1170/1283 (91.19%)\n",
      "Valid set: Average loss: 0.2688, Accuracy: 288/321 (89.72%)\n",
      "epoch= 1000: Train set: Average loss: 0.3301, Accuracy: 1165/1283 (90.80%)\n",
      "Valid set: Average loss: 0.3765, Accuracy: 263/321 (81.93%)\n",
      "epoch= 1010: Train set: Average loss: 0.2854, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.2899, Accuracy: 281/321 (87.54%)\n",
      "epoch= 1020: Train set: Average loss: 0.2703, Accuracy: 1189/1283 (92.67%)\n",
      "Valid set: Average loss: 0.2292, Accuracy: 288/321 (89.72%)\n",
      "epoch= 1030: Train set: Average loss: 0.1964, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.2279, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1040: Train set: Average loss: 0.3052, Accuracy: 1186/1283 (92.44%)\n",
      "Valid set: Average loss: 0.2247, Accuracy: 293/321 (91.28%)\n",
      "best_loss updated:  0.22468086083730063\n",
      "epoch= 1050: Train set: Average loss: 0.3441, Accuracy: 1170/1283 (91.19%)\n",
      "Valid set: Average loss: 0.2327, Accuracy: 288/321 (89.72%)\n",
      "epoch= 1060: Train set: Average loss: 0.2119, Accuracy: 1187/1283 (92.52%)\n",
      "Valid set: Average loss: 0.2238, Accuracy: 289/321 (90.03%)\n",
      "best_loss updated:  0.223779966433843\n",
      "epoch= 1070: Train set: Average loss: 0.2064, Accuracy: 1180/1283 (91.97%)\n",
      "Valid set: Average loss: 0.2253, Accuracy: 289/321 (90.03%)\n",
      "epoch= 1080: Train set: Average loss: 0.2155, Accuracy: 1188/1283 (92.60%)\n",
      "Valid set: Average loss: 0.2099, Accuracy: 294/321 (91.59%)\n",
      "best_loss updated:  0.20985362927118936\n",
      "epoch= 1090: Train set: Average loss: 0.1883, Accuracy: 1186/1283 (92.44%)\n",
      "Valid set: Average loss: 0.2077, Accuracy: 294/321 (91.59%)\n",
      "best_loss updated:  0.207687109708786\n",
      "epoch= 1100: Train set: Average loss: 0.2260, Accuracy: 1184/1283 (92.28%)\n",
      "Valid set: Average loss: 0.2217, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1110: Train set: Average loss: 0.2541, Accuracy: 1182/1283 (92.13%)\n",
      "Valid set: Average loss: 0.2287, Accuracy: 294/321 (91.59%)\n",
      "epoch= 1120: Train set: Average loss: 0.2667, Accuracy: 1177/1283 (91.74%)\n",
      "Valid set: Average loss: 0.2297, Accuracy: 289/321 (90.03%)\n",
      "epoch= 1130: Train set: Average loss: 0.2417, Accuracy: 1180/1283 (91.97%)\n",
      "Valid set: Average loss: 0.2215, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1140: Train set: Average loss: 0.1868, Accuracy: 1182/1283 (92.13%)\n",
      "Valid set: Average loss: 0.2087, Accuracy: 297/321 (92.52%)\n",
      "epoch= 1150: Train set: Average loss: 0.1932, Accuracy: 1171/1283 (91.27%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Valid set: Average loss: 0.2379, Accuracy: 285/321 (88.79%)\n",
      "epoch= 1160: Train set: Average loss: 0.1775, Accuracy: 1190/1283 (92.75%)\n",
      "Valid set: Average loss: 0.2114, Accuracy: 294/321 (91.59%)\n",
      "epoch= 1170: Train set: Average loss: 0.1745, Accuracy: 1187/1283 (92.52%)\n",
      "Valid set: Average loss: 0.2113, Accuracy: 295/321 (91.90%)\n",
      "epoch= 1180: Train set: Average loss: 0.2302, Accuracy: 1197/1283 (93.30%)\n",
      "Valid set: Average loss: 0.2118, Accuracy: 293/321 (91.28%)\n",
      "epoch= 1190: Train set: Average loss: 0.2133, Accuracy: 1168/1283 (91.04%)\n",
      "Valid set: Average loss: 0.2153, Accuracy: 291/321 (90.65%)\n",
      "epoch= 1200: Train set: Average loss: 0.2013, Accuracy: 1190/1283 (92.75%)\n",
      "Valid set: Average loss: 0.2080, Accuracy: 296/321 (92.21%)\n",
      "epoch= 1210: Train set: Average loss: 0.1791, Accuracy: 1180/1283 (91.97%)\n",
      "Valid set: Average loss: 0.2183, Accuracy: 295/321 (91.90%)\n",
      "epoch= 1220: Train set: Average loss: 0.2108, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.2153, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1230: Train set: Average loss: 0.1800, Accuracy: 1182/1283 (92.13%)\n",
      "Valid set: Average loss: 0.2246, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1240: Train set: Average loss: 0.1699, Accuracy: 1195/1283 (93.14%)\n",
      "Valid set: Average loss: 0.2177, Accuracy: 295/321 (91.90%)\n",
      "epoch= 1250: Train set: Average loss: 0.1759, Accuracy: 1192/1283 (92.91%)\n",
      "Valid set: Average loss: 0.2218, Accuracy: 294/321 (91.59%)\n",
      "epoch= 1260: Train set: Average loss: 0.1884, Accuracy: 1198/1283 (93.37%)\n",
      "Valid set: Average loss: 0.2187, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1270: Train set: Average loss: 0.2364, Accuracy: 1191/1283 (92.83%)\n",
      "Valid set: Average loss: 0.2173, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1280: Train set: Average loss: 0.3332, Accuracy: 1199/1283 (93.45%)\n",
      "Valid set: Average loss: 0.2204, Accuracy: 292/321 (90.97%)\n",
      "epoch= 1290: Train set: Average loss: 0.1704, Accuracy: 1197/1283 (93.30%)\n",
      "Valid set: Average loss: 0.2191, Accuracy: 290/321 (90.34%)\n",
      "epoch= 1300: Train set: Average loss: 0.1742, Accuracy: 1192/1283 (92.91%)\n",
      "Valid set: Average loss: 0.2167, Accuracy: 293/321 (91.28%)\n",
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EF41E71048>, <torchvision.transforms.transforms.Lambda object at 0x000002EF41E711D0>, <torchvision.transforms.transforms.ToTensor object at 0x000002EF41E71860>]\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EF41E712B0>, <torchvision.transforms.transforms.ToTensor object at 0x000002EF41E715C0>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 10: Train set: Average loss: 0.7476, Accuracy: 713/1283 (55.57%)\n",
      "Valid set: Average loss: 0.6518, Accuracy: 202/321 (62.93%)\n",
      "best_loss updated:  0.6518378456433614\n",
      "epoch= 20: Train set: Average loss: 0.5947, Accuracy: 709/1283 (55.26%)\n",
      "Valid set: Average loss: 0.6806, Accuracy: 176/321 (54.83%)\n",
      "epoch= 30: Train set: Average loss: 0.6057, Accuracy: 814/1283 (63.45%)\n",
      "Valid set: Average loss: 0.7183, Accuracy: 145/321 (45.17%)\n",
      "epoch= 40: Train set: Average loss: 0.4486, Accuracy: 1017/1283 (79.27%)\n",
      "Valid set: Average loss: 0.7309, Accuracy: 154/321 (47.98%)\n",
      "epoch= 50: Train set: Average loss: 0.6978, Accuracy: 692/1283 (53.94%)\n",
      "Valid set: Average loss: 0.6646, Accuracy: 166/321 (51.71%)\n",
      "epoch= 60: Train set: Average loss: 0.5747, Accuracy: 850/1283 (66.25%)\n",
      "Valid set: Average loss: 0.7515, Accuracy: 148/321 (46.11%)\n",
      "epoch= 70: Train set: Average loss: 0.5947, Accuracy: 919/1283 (71.63%)\n",
      "Valid set: Average loss: 0.7037, Accuracy: 169/321 (52.65%)\n",
      "epoch= 80: Train set: Average loss: 0.5013, Accuracy: 960/1283 (74.82%)\n",
      "Valid set: Average loss: 0.5660, Accuracy: 234/321 (72.90%)\n",
      "best_loss updated:  0.5660185813903809\n",
      "epoch= 90: Train set: Average loss: 0.4139, Accuracy: 1036/1283 (80.75%)\n",
      "Valid set: Average loss: 0.4845, Accuracy: 247/321 (76.95%)\n",
      "best_loss updated:  0.4845182200272878\n",
      "epoch= 100: Train set: Average loss: 0.5980, Accuracy: 856/1283 (66.72%)\n",
      "Valid set: Average loss: 0.6080, Accuracy: 201/321 (62.62%)\n",
      "epoch= 110: Train set: Average loss: 0.6309, Accuracy: 857/1283 (66.80%)\n",
      "Valid set: Average loss: 0.6598, Accuracy: 206/321 (64.17%)\n",
      "epoch= 120: Train set: Average loss: 0.4235, Accuracy: 1020/1283 (79.50%)\n",
      "Valid set: Average loss: 0.5003, Accuracy: 248/321 (77.26%)\n",
      "epoch= 130: Train set: Average loss: 0.5458, Accuracy: 992/1283 (77.32%)\n",
      "Valid set: Average loss: 0.6333, Accuracy: 216/321 (67.29%)\n",
      "epoch= 140: Train set: Average loss: 0.4591, Accuracy: 992/1283 (77.32%)\n",
      "Valid set: Average loss: 0.5842, Accuracy: 231/321 (71.96%)\n",
      "epoch= 150: Train set: Average loss: 0.5341, Accuracy: 1017/1283 (79.27%)\n",
      "Valid set: Average loss: 0.4776, Accuracy: 240/321 (74.77%)\n",
      "best_loss updated:  0.47763100266456604\n",
      "epoch= 160: Train set: Average loss: 0.4106, Accuracy: 1042/1283 (81.22%)\n",
      "Valid set: Average loss: 0.4684, Accuracy: 249/321 (77.57%)\n",
      "best_loss updated:  0.46843989690144855\n",
      "epoch= 170: Train set: Average loss: 0.4462, Accuracy: 1055/1283 (82.23%)\n",
      "Valid set: Average loss: 0.6314, Accuracy: 215/321 (66.98%)\n",
      "epoch= 180: Train set: Average loss: 0.4359, Accuracy: 1022/1283 (79.66%)\n",
      "Valid set: Average loss: 0.4998, Accuracy: 242/321 (75.39%)\n",
      "epoch= 190: Train set: Average loss: 0.5187, Accuracy: 999/1283 (77.86%)\n",
      "Valid set: Average loss: 0.5203, Accuracy: 238/321 (74.14%)\n",
      "epoch= 200: Train set: Average loss: 0.4527, Accuracy: 1043/1283 (81.29%)\n",
      "Valid set: Average loss: 0.4860, Accuracy: 248/321 (77.26%)\n",
      "epoch= 210: Train set: Average loss: 0.4269, Accuracy: 1042/1283 (81.22%)\n",
      "Valid set: Average loss: 0.4933, Accuracy: 243/321 (75.70%)\n",
      "epoch= 220: Train set: Average loss: 0.7029, Accuracy: 647/1283 (50.43%)\n",
      "Valid set: Average loss: 0.7022, Accuracy: 172/321 (53.58%)\n",
      "epoch= 230: Train set: Average loss: 0.6320, Accuracy: 839/1283 (65.39%)\n",
      "Valid set: Average loss: 0.6156, Accuracy: 220/321 (68.54%)\n",
      "epoch= 240: Train set: Average loss: 0.6157, Accuracy: 867/1283 (67.58%)\n",
      "Valid set: Average loss: 1.4763, Accuracy: 173/321 (53.89%)\n",
      "epoch= 250: Train set: Average loss: 0.4400, Accuracy: 1020/1283 (79.50%)\n",
      "Valid set: Average loss: 0.4895, Accuracy: 248/321 (77.26%)\n",
      "epoch= 260: Train set: Average loss: 0.4556, Accuracy: 1028/1283 (80.12%)\n",
      "Valid set: Average loss: 0.4705, Accuracy: 252/321 (78.50%)\n",
      "epoch= 270: Train set: Average loss: 0.3818, Accuracy: 1067/1283 (83.16%)\n",
      "Valid set: Average loss: 0.4936, Accuracy: 245/321 (76.32%)\n",
      "epoch= 280: Train set: Average loss: 0.5076, Accuracy: 1035/1283 (80.67%)\n",
      "Valid set: Average loss: 0.4954, Accuracy: 250/321 (77.88%)\n",
      "epoch= 290: Train set: Average loss: 0.4158, Accuracy: 1048/1283 (81.68%)\n",
      "Valid set: Average loss: 0.4729, Accuracy: 252/321 (78.50%)\n",
      "epoch= 300: Train set: Average loss: 0.4369, Accuracy: 1057/1283 (82.39%)\n",
      "Valid set: Average loss: 0.4897, Accuracy: 251/321 (78.19%)\n",
      "epoch= 310: Train set: Average loss: 0.4031, Accuracy: 1060/1283 (82.62%)\n",
      "Valid set: Average loss: 0.4623, Accuracy: 254/321 (79.13%)\n",
      "best_loss updated:  0.4623004198074341\n",
      "epoch= 320: Train set: Average loss: 0.3844, Accuracy: 1060/1283 (82.62%)\n",
      "Valid set: Average loss: 0.4749, Accuracy: 251/321 (78.19%)\n",
      "epoch= 330: Train set: Average loss: 0.3631, Accuracy: 1083/1283 (84.41%)\n",
      "Valid set: Average loss: 0.4521, Accuracy: 256/321 (79.75%)\n",
      "best_loss updated:  0.45208917061487836\n",
      "epoch= 340: Train set: Average loss: 0.3596, Accuracy: 1084/1283 (84.49%)\n",
      "Valid set: Average loss: 0.4445, Accuracy: 252/321 (78.50%)\n",
      "best_loss updated:  0.44450415174166363\n",
      "epoch= 350: Train set: Average loss: 0.4470, Accuracy: 1098/1283 (85.58%)\n",
      "Valid set: Average loss: 0.4503, Accuracy: 255/321 (79.44%)\n",
      "epoch= 360: Train set: Average loss: 0.3820, Accuracy: 1067/1283 (83.16%)\n",
      "Valid set: Average loss: 0.4465, Accuracy: 250/321 (77.88%)\n",
      "epoch= 370: Train set: Average loss: 0.3481, Accuracy: 1093/1283 (85.19%)\n",
      "Valid set: Average loss: 0.4576, Accuracy: 248/321 (77.26%)\n",
      "epoch= 380: Train set: Average loss: 0.3519, Accuracy: 1076/1283 (83.87%)\n",
      "Valid set: Average loss: 0.4714, Accuracy: 249/321 (77.57%)\n",
      "epoch= 390: Train set: Average loss: 0.3202, Accuracy: 1090/1283 (84.96%)\n",
      "Valid set: Average loss: 0.3715, Accuracy: 263/321 (81.93%)\n",
      "best_loss updated:  0.3715486228466034\n",
      "epoch= 400: Train set: Average loss: 0.3672, Accuracy: 1084/1283 (84.49%)\n",
      "Valid set: Average loss: 0.3865, Accuracy: 262/321 (81.62%)\n",
      "epoch= 410: Train set: Average loss: 0.3482, Accuracy: 1093/1283 (85.19%)\n",
      "Valid set: Average loss: 0.3896, Accuracy: 266/321 (82.87%)\n",
      "epoch= 420: Train set: Average loss: 0.4390, Accuracy: 1090/1283 (84.96%)\n",
      "Valid set: Average loss: 0.4161, Accuracy: 260/321 (81.00%)\n",
      "epoch= 430: Train set: Average loss: 0.3053, Accuracy: 1094/1283 (85.27%)\n",
      "Valid set: Average loss: 0.3906, Accuracy: 261/321 (81.31%)\n",
      "epoch= 440: Train set: Average loss: 0.3282, Accuracy: 1118/1283 (87.14%)\n",
      "Valid set: Average loss: 0.4395, Accuracy: 253/321 (78.82%)\n",
      "epoch= 450: Train set: Average loss: 0.3081, Accuracy: 1094/1283 (85.27%)\n",
      "Valid set: Average loss: 0.3962, Accuracy: 261/321 (81.31%)\n",
      "epoch= 460: Train set: Average loss: 0.2937, Accuracy: 1117/1283 (87.06%)\n",
      "Valid set: Average loss: 0.3767, Accuracy: 266/321 (82.87%)\n",
      "epoch= 470: Train set: Average loss: 0.3152, Accuracy: 1094/1283 (85.27%)\n",
      "Valid set: Average loss: 0.3573, Accuracy: 267/321 (83.18%)\n",
      "best_loss updated:  0.35730058948198956\n",
      "epoch= 480: Train set: Average loss: 0.3854, Accuracy: 1099/1283 (85.66%)\n",
      "Valid set: Average loss: 0.3900, Accuracy: 261/321 (81.31%)\n",
      "epoch= 490: Train set: Average loss: 0.2915, Accuracy: 1109/1283 (86.44%)\n",
      "Valid set: Average loss: 0.3508, Accuracy: 266/321 (82.87%)\n",
      "best_loss updated:  0.35083480676015216\n",
      "epoch= 500: Train set: Average loss: 0.2810, Accuracy: 1110/1283 (86.52%)\n",
      "Valid set: Average loss: 0.3562, Accuracy: 265/321 (82.55%)\n",
      "epoch= 510: Train set: Average loss: 0.3155, Accuracy: 1125/1283 (87.69%)\n",
      "Valid set: Average loss: 0.3556, Accuracy: 269/321 (83.80%)\n",
      "epoch= 520: Train set: Average loss: 0.2772, Accuracy: 1115/1283 (86.91%)\n",
      "Valid set: Average loss: 0.3181, Accuracy: 269/321 (83.80%)\n",
      "best_loss updated:  0.318084975083669\n",
      "epoch= 530: Train set: Average loss: 0.2933, Accuracy: 1135/1283 (88.46%)\n",
      "Valid set: Average loss: 0.3470, Accuracy: 270/321 (84.11%)\n",
      "epoch= 540: Train set: Average loss: 0.3559, Accuracy: 1125/1283 (87.69%)\n",
      "Valid set: Average loss: 0.2982, Accuracy: 274/321 (85.36%)\n",
      "best_loss updated:  0.2982305884361267\n",
      "epoch= 550: Train set: Average loss: 0.2877, Accuracy: 1119/1283 (87.22%)\n",
      "Valid set: Average loss: 0.3726, Accuracy: 263/321 (81.93%)\n",
      "epoch= 560: Train set: Average loss: 0.3049, Accuracy: 1108/1283 (86.36%)\n",
      "Valid set: Average loss: 0.3983, Accuracy: 256/321 (79.75%)\n",
      "epoch= 570: Train set: Average loss: 0.4264, Accuracy: 1107/1283 (86.28%)\n",
      "Valid set: Average loss: 0.3185, Accuracy: 269/321 (83.80%)\n",
      "epoch= 580: Train set: Average loss: 0.3135, Accuracy: 1114/1283 (86.83%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Valid set: Average loss: 0.3231, Accuracy: 266/321 (82.87%)\n",
      "epoch= 590: Train set: Average loss: 0.2790, Accuracy: 1132/1283 (88.23%)\n",
      "Valid set: Average loss: 0.3544, Accuracy: 264/321 (82.24%)\n",
      "epoch= 600: Train set: Average loss: 0.3399, Accuracy: 1083/1283 (84.41%)\n",
      "Valid set: Average loss: 0.4876, Accuracy: 234/321 (72.90%)\n",
      "epoch= 610: Train set: Average loss: 0.3387, Accuracy: 1111/1283 (86.59%)\n",
      "Valid set: Average loss: 0.3576, Accuracy: 267/321 (83.18%)\n",
      "epoch= 620: Train set: Average loss: 0.3091, Accuracy: 1126/1283 (87.76%)\n",
      "Valid set: Average loss: 0.3406, Accuracy: 262/321 (81.62%)\n",
      "epoch= 630: Train set: Average loss: 0.3185, Accuracy: 1142/1283 (89.01%)\n",
      "Valid set: Average loss: 0.3120, Accuracy: 272/321 (84.74%)\n",
      "epoch= 640: Train set: Average loss: 0.2917, Accuracy: 1120/1283 (87.30%)\n",
      "Valid set: Average loss: 0.5211, Accuracy: 231/321 (71.96%)\n",
      "epoch= 650: Train set: Average loss: 0.3718, Accuracy: 1104/1283 (86.05%)\n",
      "Valid set: Average loss: 0.3218, Accuracy: 273/321 (85.05%)\n",
      "epoch= 660: Train set: Average loss: 0.3403, Accuracy: 1132/1283 (88.23%)\n",
      "Valid set: Average loss: 0.2584, Accuracy: 279/321 (86.92%)\n",
      "best_loss updated:  0.25839315354824066\n",
      "epoch= 670: Train set: Average loss: 0.3919, Accuracy: 1128/1283 (87.92%)\n",
      "Valid set: Average loss: 0.3120, Accuracy: 271/321 (84.42%)\n",
      "epoch= 680: Train set: Average loss: 0.2358, Accuracy: 1152/1283 (89.79%)\n",
      "Valid set: Average loss: 0.2962, Accuracy: 265/321 (82.55%)\n",
      "epoch= 690: Train set: Average loss: 0.2735, Accuracy: 1121/1283 (87.37%)\n",
      "Valid set: Average loss: 0.2932, Accuracy: 271/321 (84.42%)\n",
      "epoch= 700: Train set: Average loss: 0.3187, Accuracy: 1123/1283 (87.53%)\n",
      "Valid set: Average loss: 0.3707, Accuracy: 266/321 (82.87%)\n",
      "epoch= 710: Train set: Average loss: 0.4268, Accuracy: 1131/1283 (88.15%)\n",
      "Valid set: Average loss: 0.3447, Accuracy: 264/321 (82.24%)\n",
      "epoch= 720: Train set: Average loss: 0.2882, Accuracy: 1139/1283 (88.78%)\n",
      "Valid set: Average loss: 0.3260, Accuracy: 269/321 (83.80%)\n",
      "epoch= 730: Train set: Average loss: 0.2710, Accuracy: 1126/1283 (87.76%)\n",
      "Valid set: Average loss: 0.3565, Accuracy: 266/321 (82.87%)\n",
      "epoch= 740: Train set: Average loss: 0.2831, Accuracy: 1110/1283 (86.52%)\n",
      "Valid set: Average loss: 0.3628, Accuracy: 266/321 (82.87%)\n",
      "epoch= 750: Train set: Average loss: 0.2766, Accuracy: 1130/1283 (88.07%)\n",
      "Valid set: Average loss: 0.3202, Accuracy: 267/321 (83.18%)\n",
      "epoch= 760: Train set: Average loss: 0.2523, Accuracy: 1137/1283 (88.62%)\n",
      "Valid set: Average loss: 0.2696, Accuracy: 278/321 (86.60%)\n",
      "epoch= 770: Train set: Average loss: 0.3146, Accuracy: 1146/1283 (89.32%)\n",
      "Valid set: Average loss: 0.2737, Accuracy: 278/321 (86.60%)\n",
      "epoch= 780: Train set: Average loss: 0.3173, Accuracy: 1086/1283 (84.65%)\n",
      "Valid set: Average loss: 0.3113, Accuracy: 279/321 (86.92%)\n",
      "epoch= 790: Train set: Average loss: 0.3277, Accuracy: 1146/1283 (89.32%)\n",
      "Valid set: Average loss: 0.3307, Accuracy: 270/321 (84.11%)\n",
      "epoch= 800: Train set: Average loss: 0.3417, Accuracy: 1131/1283 (88.15%)\n",
      "Valid set: Average loss: 0.3225, Accuracy: 269/321 (83.80%)\n",
      "epoch= 810: Train set: Average loss: 0.2409, Accuracy: 1135/1283 (88.46%)\n",
      "Valid set: Average loss: 0.2767, Accuracy: 278/321 (86.60%)\n",
      "epoch= 820: Train set: Average loss: 0.2533, Accuracy: 1144/1283 (89.17%)\n",
      "Valid set: Average loss: 0.2997, Accuracy: 269/321 (83.80%)\n",
      "epoch= 830: Train set: Average loss: 0.2327, Accuracy: 1159/1283 (90.34%)\n",
      "Valid set: Average loss: 0.2910, Accuracy: 274/321 (85.36%)\n",
      "epoch= 840: Train set: Average loss: 0.2640, Accuracy: 1138/1283 (88.70%)\n",
      "Valid set: Average loss: 0.2632, Accuracy: 280/321 (87.23%)\n",
      "epoch= 850: Train set: Average loss: 0.2320, Accuracy: 1153/1283 (89.87%)\n",
      "Valid set: Average loss: 0.2579, Accuracy: 276/321 (85.98%)\n",
      "best_loss updated:  0.2579068640867869\n",
      "epoch= 860: Train set: Average loss: 0.2923, Accuracy: 1123/1283 (87.53%)\n",
      "Valid set: Average loss: 0.2699, Accuracy: 279/321 (86.92%)\n",
      "epoch= 870: Train set: Average loss: 0.2536, Accuracy: 1155/1283 (90.02%)\n",
      "Valid set: Average loss: 0.2721, Accuracy: 280/321 (87.23%)\n",
      "epoch= 880: Train set: Average loss: 0.3525, Accuracy: 1131/1283 (88.15%)\n",
      "Valid set: Average loss: 0.2670, Accuracy: 280/321 (87.23%)\n",
      "epoch= 890: Train set: Average loss: 0.2275, Accuracy: 1161/1283 (90.49%)\n",
      "Valid set: Average loss: 0.2439, Accuracy: 281/321 (87.54%)\n",
      "best_loss updated:  0.2439310004313787\n",
      "epoch= 900: Train set: Average loss: 0.2595, Accuracy: 1139/1283 (88.78%)\n",
      "Valid set: Average loss: 0.6290, Accuracy: 237/321 (73.83%)\n",
      "epoch= 910: Train set: Average loss: 0.3465, Accuracy: 1155/1283 (90.02%)\n",
      "Valid set: Average loss: 0.4166, Accuracy: 259/321 (80.69%)\n",
      "epoch= 920: Train set: Average loss: 0.3432, Accuracy: 1140/1283 (88.85%)\n",
      "Valid set: Average loss: 0.2718, Accuracy: 287/321 (89.41%)\n",
      "epoch= 930: Train set: Average loss: 0.2233, Accuracy: 1155/1283 (90.02%)\n",
      "Valid set: Average loss: 0.2685, Accuracy: 280/321 (87.23%)\n",
      "epoch= 940: Train set: Average loss: 0.3063, Accuracy: 1115/1283 (86.91%)\n",
      "Valid set: Average loss: 0.6523, Accuracy: 211/321 (65.73%)\n",
      "epoch= 950: Train set: Average loss: 0.2434, Accuracy: 1143/1283 (89.09%)\n",
      "Valid set: Average loss: 0.3231, Accuracy: 273/321 (85.05%)\n",
      "epoch= 960: Train set: Average loss: 0.2960, Accuracy: 1135/1283 (88.46%)\n",
      "Valid set: Average loss: 0.2804, Accuracy: 277/321 (86.29%)\n",
      "epoch= 970: Train set: Average loss: 0.3583, Accuracy: 1133/1283 (88.31%)\n",
      "Valid set: Average loss: 0.2977, Accuracy: 278/321 (86.60%)\n",
      "epoch= 980: Train set: Average loss: 0.2039, Accuracy: 1162/1283 (90.57%)\n",
      "Valid set: Average loss: 0.2437, Accuracy: 280/321 (87.23%)\n",
      "best_loss updated:  0.24374557038148245\n",
      "epoch= 990: Train set: Average loss: 0.2728, Accuracy: 1161/1283 (90.49%)\n",
      "Valid set: Average loss: 0.2622, Accuracy: 284/321 (88.47%)\n",
      "epoch= 1000: Train set: Average loss: 0.2414, Accuracy: 1158/1283 (90.26%)\n",
      "Valid set: Average loss: 0.2206, Accuracy: 290/321 (90.34%)\n",
      "best_loss updated:  0.22061569492022196\n",
      "epoch= 1010: Train set: Average loss: 0.3161, Accuracy: 1165/1283 (90.80%)\n",
      "Valid set: Average loss: 0.2199, Accuracy: 286/321 (89.10%)\n",
      "best_loss updated:  0.2198925664027532\n",
      "epoch= 1020: Train set: Average loss: 0.2378, Accuracy: 1164/1283 (90.72%)\n",
      "Valid set: Average loss: 0.2416, Accuracy: 285/321 (88.79%)\n",
      "epoch= 1030: Train set: Average loss: 0.2876, Accuracy: 1179/1283 (91.89%)\n",
      "Valid set: Average loss: 0.2388, Accuracy: 283/321 (88.16%)\n",
      "epoch= 1040: Train set: Average loss: 0.3552, Accuracy: 1166/1283 (90.88%)\n",
      "Valid set: Average loss: 0.2704, Accuracy: 276/321 (85.98%)\n",
      "epoch= 1050: Train set: Average loss: 0.1847, Accuracy: 1175/1283 (91.58%)\n",
      "Valid set: Average loss: 0.2479, Accuracy: 282/321 (87.85%)\n",
      "epoch= 1060: Train set: Average loss: 0.2063, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.2239, Accuracy: 284/321 (88.47%)\n",
      "epoch= 1070: Train set: Average loss: 0.2638, Accuracy: 1173/1283 (91.43%)\n",
      "Valid set: Average loss: 0.2539, Accuracy: 282/321 (87.85%)\n",
      "epoch= 1080: Train set: Average loss: 0.2592, Accuracy: 1163/1283 (90.65%)\n",
      "Valid set: Average loss: 0.2525, Accuracy: 284/321 (88.47%)\n",
      "epoch= 1090: Train set: Average loss: 0.2458, Accuracy: 1176/1283 (91.66%)\n",
      "Valid set: Average loss: 0.2639, Accuracy: 283/321 (88.16%)\n",
      "epoch= 1100: Train set: Average loss: 0.2616, Accuracy: 1153/1283 (89.87%)\n",
      "Valid set: Average loss: 0.2465, Accuracy: 283/321 (88.16%)\n",
      "epoch= 1110: Train set: Average loss: 0.2363, Accuracy: 1163/1283 (90.65%)\n",
      "Valid set: Average loss: 0.2427, Accuracy: 288/321 (89.72%)\n",
      "epoch= 1120: Train set: Average loss: 0.2077, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.2336, Accuracy: 285/321 (88.79%)\n",
      "epoch= 1130: Train set: Average loss: 0.1873, Accuracy: 1173/1283 (91.43%)\n",
      "Valid set: Average loss: 0.2316, Accuracy: 285/321 (88.79%)\n",
      "epoch= 1140: Train set: Average loss: 0.1927, Accuracy: 1180/1283 (91.97%)\n",
      "Valid set: Average loss: 0.2404, Accuracy: 282/321 (87.85%)\n",
      "epoch= 1150: Train set: Average loss: 0.2375, Accuracy: 1181/1283 (92.05%)\n",
      "Valid set: Average loss: 0.2380, Accuracy: 287/321 (89.41%)\n",
      "epoch= 1160: Train set: Average loss: 0.2339, Accuracy: 1172/1283 (91.35%)\n",
      "Valid set: Average loss: 0.2314, Accuracy: 284/321 (88.47%)\n",
      "epoch= 1170: Train set: Average loss: 0.1993, Accuracy: 1176/1283 (91.66%)\n",
      "Valid set: Average loss: 0.2568, Accuracy: 281/321 (87.54%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 1180: Train set: Average loss: 0.1844, Accuracy: 1179/1283 (91.89%)\n",
      "Valid set: Average loss: 0.2452, Accuracy: 282/321 (87.85%)\n",
      "epoch= 1190: Train set: Average loss: 0.2441, Accuracy: 1169/1283 (91.11%)\n",
      "Valid set: Average loss: 0.2425, Accuracy: 287/321 (89.41%)\n",
      "epoch= 1200: Train set: Average loss: 0.1884, Accuracy: 1177/1283 (91.74%)\n",
      "Valid set: Average loss: 0.2377, Accuracy: 286/321 (89.10%)\n",
      "epoch= 1210: Train set: Average loss: 0.3578, Accuracy: 1171/1283 (91.27%)\n",
      "Valid set: Average loss: 0.2488, Accuracy: 283/321 (88.16%)\n",
      "epoch= 1220: Train set: Average loss: 0.2153, Accuracy: 1170/1283 (91.19%)\n",
      "Valid set: Average loss: 0.2445, Accuracy: 283/321 (88.16%)\n",
      "epoch= 1230: Train set: Average loss: 0.1906, Accuracy: 1180/1283 (91.97%)\n",
      "Valid set: Average loss: 0.2366, Accuracy: 285/321 (88.79%)\n",
      "epoch= 1240: Train set: Average loss: 0.2063, Accuracy: 1175/1283 (91.58%)\n",
      "Valid set: Average loss: 0.2254, Accuracy: 287/321 (89.41%)\n",
      "epoch= 1250: Train set: Average loss: 0.2070, Accuracy: 1182/1283 (92.13%)\n",
      "Valid set: Average loss: 0.2400, Accuracy: 286/321 (89.10%)\n",
      "epoch= 1260: Train set: Average loss: 0.2365, Accuracy: 1185/1283 (92.36%)\n",
      "Valid set: Average loss: 0.2284, Accuracy: 286/321 (89.10%)\n",
      "epoch= 1270: Train set: Average loss: 0.2763, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.2427, Accuracy: 284/321 (88.47%)\n",
      "epoch= 1280: Train set: Average loss: 0.1985, Accuracy: 1184/1283 (92.28%)\n",
      "Valid set: Average loss: 0.2442, Accuracy: 286/321 (89.10%)\n",
      "epoch= 1290: Train set: Average loss: 0.2274, Accuracy: 1187/1283 (92.52%)\n",
      "Valid set: Average loss: 0.2365, Accuracy: 284/321 (88.47%)\n",
      "epoch= 1300: Train set: Average loss: 0.2759, Accuracy: 1174/1283 (91.50%)\n",
      "Valid set: Average loss: 0.2286, Accuracy: 285/321 (88.79%)\n",
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EF20C7E2E8>, <torchvision.transforms.transforms.Lambda object at 0x000002EF20CA5CC0>, <torchvision.transforms.transforms.ToTensor object at 0x000002EF20CA5BE0>]\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EF20CA5E48>, <torchvision.transforms.transforms.ToTensor object at 0x000002EF20CA5438>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 10: Train set: Average loss: 1.2655, Accuracy: 804/1284 (62.62%)\n",
      "Valid set: Average loss: 0.6310, Accuracy: 201/320 (62.81%)\n",
      "best_loss updated:  0.6310044129689535\n",
      "epoch= 20: Train set: Average loss: 0.7244, Accuracy: 665/1284 (51.79%)\n",
      "Valid set: Average loss: 0.6914, Accuracy: 150/320 (46.88%)\n",
      "epoch= 30: Train set: Average loss: 0.6007, Accuracy: 789/1284 (61.45%)\n",
      "Valid set: Average loss: 0.6667, Accuracy: 190/320 (59.38%)\n",
      "epoch= 40: Train set: Average loss: 0.6125, Accuracy: 852/1284 (66.36%)\n",
      "Valid set: Average loss: 0.7072, Accuracy: 169/320 (52.81%)\n",
      "epoch= 50: Train set: Average loss: 0.5264, Accuracy: 914/1284 (71.18%)\n",
      "Valid set: Average loss: 0.6000, Accuracy: 210/320 (65.62%)\n",
      "best_loss updated:  0.5999535719553629\n",
      "epoch= 60: Train set: Average loss: 0.4934, Accuracy: 967/1284 (75.31%)\n",
      "Valid set: Average loss: 0.5344, Accuracy: 249/320 (77.81%)\n",
      "best_loss updated:  0.5344193975130717\n",
      "epoch= 70: Train set: Average loss: 0.5081, Accuracy: 975/1284 (75.93%)\n",
      "Valid set: Average loss: 0.4239, Accuracy: 258/320 (80.62%)\n",
      "best_loss updated:  0.42387985189755756\n",
      "epoch= 80: Train set: Average loss: 0.4710, Accuracy: 974/1284 (75.86%)\n",
      "Valid set: Average loss: 0.4822, Accuracy: 251/320 (78.44%)\n",
      "epoch= 90: Train set: Average loss: 0.5407, Accuracy: 990/1284 (77.10%)\n",
      "Valid set: Average loss: 0.5929, Accuracy: 227/320 (70.94%)\n",
      "epoch= 100: Train set: Average loss: 0.4414, Accuracy: 1033/1284 (80.45%)\n",
      "Valid set: Average loss: 0.6092, Accuracy: 163/320 (50.94%)\n",
      "epoch= 110: Train set: Average loss: 0.4293, Accuracy: 1031/1284 (80.30%)\n",
      "Valid set: Average loss: 0.4349, Accuracy: 270/320 (84.38%)\n",
      "epoch= 120: Train set: Average loss: 0.4570, Accuracy: 1057/1284 (82.32%)\n",
      "Valid set: Average loss: 1.5178, Accuracy: 163/320 (50.94%)\n",
      "epoch= 130: Train set: Average loss: 0.4199, Accuracy: 1044/1284 (81.31%)\n",
      "Valid set: Average loss: 0.4099, Accuracy: 268/320 (83.75%)\n",
      "best_loss updated:  0.40988993644714355\n",
      "epoch= 140: Train set: Average loss: 0.4181, Accuracy: 1030/1284 (80.22%)\n",
      "Valid set: Average loss: 0.4124, Accuracy: 264/320 (82.50%)\n",
      "epoch= 150: Train set: Average loss: 0.4334, Accuracy: 1046/1284 (81.46%)\n",
      "Valid set: Average loss: 0.4177, Accuracy: 265/320 (82.81%)\n",
      "epoch= 160: Train set: Average loss: 0.4652, Accuracy: 1014/1284 (78.97%)\n",
      "Valid set: Average loss: 0.4659, Accuracy: 246/320 (76.88%)\n",
      "epoch= 170: Train set: Average loss: 0.3693, Accuracy: 1074/1284 (83.64%)\n",
      "Valid set: Average loss: 0.3688, Accuracy: 268/320 (83.75%)\n",
      "best_loss updated:  0.36876265207926434\n",
      "epoch= 180: Train set: Average loss: 0.4038, Accuracy: 1045/1284 (81.39%)\n",
      "Valid set: Average loss: 0.5401, Accuracy: 239/320 (74.69%)\n",
      "epoch= 190: Train set: Average loss: 0.4803, Accuracy: 990/1284 (77.10%)\n",
      "Valid set: Average loss: 0.5284, Accuracy: 240/320 (75.00%)\n",
      "epoch= 200: Train set: Average loss: 0.4611, Accuracy: 1044/1284 (81.31%)\n",
      "Valid set: Average loss: 0.4160, Accuracy: 270/320 (84.38%)\n",
      "epoch= 210: Train set: Average loss: 0.3937, Accuracy: 1047/1284 (81.54%)\n",
      "Valid set: Average loss: 0.4669, Accuracy: 254/320 (79.38%)\n",
      "epoch= 220: Train set: Average loss: 0.4007, Accuracy: 1044/1284 (81.31%)\n",
      "Valid set: Average loss: 0.5242, Accuracy: 237/320 (74.06%)\n",
      "epoch= 230: Train set: Average loss: 0.3812, Accuracy: 1078/1284 (83.96%)\n",
      "Valid set: Average loss: 0.4975, Accuracy: 258/320 (80.62%)\n",
      "epoch= 240: Train set: Average loss: 0.4792, Accuracy: 1076/1284 (83.80%)\n",
      "Valid set: Average loss: 0.3354, Accuracy: 270/320 (84.38%)\n",
      "best_loss updated:  0.33539050817489624\n",
      "epoch= 250: Train set: Average loss: 0.3833, Accuracy: 1084/1284 (84.42%)\n",
      "Valid set: Average loss: 0.5566, Accuracy: 248/320 (77.50%)\n",
      "epoch= 260: Train set: Average loss: 0.4247, Accuracy: 1086/1284 (84.58%)\n",
      "Valid set: Average loss: 0.3734, Accuracy: 262/320 (81.88%)\n",
      "epoch= 270: Train set: Average loss: 0.4560, Accuracy: 1061/1284 (82.63%)\n",
      "Valid set: Average loss: 2.7090, Accuracy: 163/320 (50.94%)\n",
      "epoch= 280: Train set: Average loss: 0.3273, Accuracy: 1087/1284 (84.66%)\n",
      "Valid set: Average loss: 0.3361, Accuracy: 275/320 (85.94%)\n",
      "epoch= 290: Train set: Average loss: 0.3682, Accuracy: 1099/1284 (85.59%)\n",
      "Valid set: Average loss: 0.3102, Accuracy: 273/320 (85.31%)\n",
      "best_loss updated:  0.3101616899172465\n",
      "epoch= 300: Train set: Average loss: 0.4070, Accuracy: 1081/1284 (84.19%)\n",
      "Valid set: Average loss: 0.4678, Accuracy: 259/320 (80.94%)\n",
      "epoch= 310: Train set: Average loss: 0.2967, Accuracy: 1133/1284 (88.24%)\n",
      "Valid set: Average loss: 0.3468, Accuracy: 282/320 (88.12%)\n",
      "epoch= 320: Train set: Average loss: 0.3583, Accuracy: 1088/1284 (84.74%)\n",
      "Valid set: Average loss: 0.2950, Accuracy: 286/320 (89.38%)\n",
      "best_loss updated:  0.29504120349884033\n",
      "epoch= 330: Train set: Average loss: 0.3230, Accuracy: 1125/1284 (87.62%)\n",
      "Valid set: Average loss: 0.4102, Accuracy: 255/320 (79.69%)\n",
      "epoch= 340: Train set: Average loss: 0.3029, Accuracy: 1136/1284 (88.47%)\n",
      "Valid set: Average loss: 0.3141, Accuracy: 271/320 (84.69%)\n",
      "epoch= 350: Train set: Average loss: 0.2845, Accuracy: 1149/1284 (89.49%)\n",
      "Valid set: Average loss: 0.3588, Accuracy: 264/320 (82.50%)\n",
      "epoch= 360: Train set: Average loss: 0.2574, Accuracy: 1139/1284 (88.71%)\n",
      "Valid set: Average loss: 0.2699, Accuracy: 283/320 (88.44%)\n",
      "best_loss updated:  0.2698771506547928\n",
      "epoch= 370: Train set: Average loss: 0.2233, Accuracy: 1162/1284 (90.50%)\n",
      "Valid set: Average loss: 0.2585, Accuracy: 293/320 (91.56%)\n",
      "best_loss updated:  0.2584538956483205\n",
      "epoch= 380: Train set: Average loss: 0.2380, Accuracy: 1160/1284 (90.34%)\n",
      "Valid set: Average loss: 0.2443, Accuracy: 293/320 (91.56%)\n",
      "best_loss updated:  0.24425759414831796\n",
      "epoch= 390: Train set: Average loss: 0.2802, Accuracy: 1127/1284 (87.77%)\n",
      "Valid set: Average loss: 0.8445, Accuracy: 210/320 (65.62%)\n",
      "epoch= 400: Train set: Average loss: 0.3051, Accuracy: 1140/1284 (88.79%)\n",
      "Valid set: Average loss: 0.3348, Accuracy: 270/320 (84.38%)\n",
      "epoch= 410: Train set: Average loss: 0.2853, Accuracy: 1137/1284 (88.55%)\n",
      "Valid set: Average loss: 0.3132, Accuracy: 286/320 (89.38%)\n",
      "epoch= 420: Train set: Average loss: 0.2349, Accuracy: 1146/1284 (89.25%)\n",
      "Valid set: Average loss: 0.4166, Accuracy: 264/320 (82.50%)\n",
      "epoch= 430: Train set: Average loss: 0.2387, Accuracy: 1154/1284 (89.88%)\n",
      "Valid set: Average loss: 0.3980, Accuracy: 283/320 (88.44%)\n",
      "epoch= 440: Train set: Average loss: 0.2100, Accuracy: 1172/1284 (91.28%)\n",
      "Valid set: Average loss: 0.3231, Accuracy: 279/320 (87.19%)\n",
      "epoch= 450: Train set: Average loss: 0.2760, Accuracy: 1119/1284 (87.15%)\n",
      "Valid set: Average loss: 0.4899, Accuracy: 255/320 (79.69%)\n",
      "epoch= 460: Train set: Average loss: 0.2546, Accuracy: 1148/1284 (89.41%)\n",
      "Valid set: Average loss: 0.2631, Accuracy: 282/320 (88.12%)\n",
      "epoch= 470: Train set: Average loss: 0.3396, Accuracy: 1157/1284 (90.11%)\n",
      "Valid set: Average loss: 0.2461, Accuracy: 288/320 (90.00%)\n",
      "epoch= 480: Train set: Average loss: 0.2250, Accuracy: 1162/1284 (90.50%)\n",
      "Valid set: Average loss: 0.3524, Accuracy: 281/320 (87.81%)\n",
      "epoch= 490: Train set: Average loss: 0.2550, Accuracy: 1164/1284 (90.65%)\n",
      "Valid set: Average loss: 0.2487, Accuracy: 293/320 (91.56%)\n",
      "epoch= 500: Train set: Average loss: 0.2621, Accuracy: 1156/1284 (90.03%)\n",
      "Valid set: Average loss: 0.2359, Accuracy: 290/320 (90.62%)\n",
      "best_loss updated:  0.23589218656222025\n",
      "epoch= 510: Train set: Average loss: 0.2608, Accuracy: 1171/1284 (91.20%)\n",
      "Valid set: Average loss: 0.2649, Accuracy: 290/320 (90.62%)\n",
      "epoch= 520: Train set: Average loss: 0.3269, Accuracy: 1165/1284 (90.73%)\n",
      "Valid set: Average loss: 0.3044, Accuracy: 284/320 (88.75%)\n",
      "epoch= 530: Train set: Average loss: 0.2558, Accuracy: 1161/1284 (90.42%)\n",
      "Valid set: Average loss: 0.3358, Accuracy: 277/320 (86.56%)\n",
      "epoch= 540: Train set: Average loss: 0.2077, Accuracy: 1167/1284 (90.89%)\n",
      "Valid set: Average loss: 0.3044, Accuracy: 279/320 (87.19%)\n",
      "epoch= 550: Train set: Average loss: 0.2319, Accuracy: 1173/1284 (91.36%)\n",
      "Valid set: Average loss: 0.2501, Accuracy: 291/320 (90.94%)\n",
      "epoch= 560: Train set: Average loss: 0.2426, Accuracy: 1150/1284 (89.56%)\n",
      "Valid set: Average loss: 0.2945, Accuracy: 278/320 (86.88%)\n",
      "epoch= 570: Train set: Average loss: 0.3064, Accuracy: 1149/1284 (89.49%)\n",
      "Valid set: Average loss: 0.4833, Accuracy: 275/320 (85.94%)\n",
      "epoch= 580: Train set: Average loss: 0.2203, Accuracy: 1168/1284 (90.97%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Valid set: Average loss: 0.2336, Accuracy: 291/320 (90.94%)\n",
      "best_loss updated:  0.2336197296778361\n",
      "epoch= 590: Train set: Average loss: 0.2859, Accuracy: 1124/1284 (87.54%)\n",
      "Valid set: Average loss: 0.6396, Accuracy: 249/320 (77.81%)\n",
      "epoch= 600: Train set: Average loss: 0.3290, Accuracy: 1168/1284 (90.97%)\n",
      "Valid set: Average loss: 0.2288, Accuracy: 291/320 (90.94%)\n",
      "best_loss updated:  0.22880076368649802\n",
      "epoch= 610: Train set: Average loss: 0.2148, Accuracy: 1163/1284 (90.58%)\n",
      "Valid set: Average loss: 0.3098, Accuracy: 284/320 (88.75%)\n",
      "epoch= 620: Train set: Average loss: 0.2802, Accuracy: 1179/1284 (91.82%)\n",
      "Valid set: Average loss: 0.6449, Accuracy: 244/320 (76.25%)\n",
      "epoch= 630: Train set: Average loss: 0.2881, Accuracy: 1173/1284 (91.36%)\n",
      "Valid set: Average loss: 1.8257, Accuracy: 179/320 (55.94%)\n",
      "epoch= 640: Train set: Average loss: 0.2145, Accuracy: 1171/1284 (91.20%)\n",
      "Valid set: Average loss: 0.3377, Accuracy: 274/320 (85.62%)\n",
      "epoch= 650: Train set: Average loss: 0.2029, Accuracy: 1168/1284 (90.97%)\n",
      "Valid set: Average loss: 0.4715, Accuracy: 256/320 (80.00%)\n",
      "epoch= 660: Train set: Average loss: 0.2162, Accuracy: 1158/1284 (90.19%)\n",
      "Valid set: Average loss: 0.2938, Accuracy: 289/320 (90.31%)\n",
      "epoch= 670: Train set: Average loss: 0.2478, Accuracy: 1182/1284 (92.06%)\n",
      "Valid set: Average loss: 0.3050, Accuracy: 287/320 (89.69%)\n",
      "epoch= 680: Train set: Average loss: 0.2185, Accuracy: 1167/1284 (90.89%)\n",
      "Valid set: Average loss: 0.2748, Accuracy: 284/320 (88.75%)\n",
      "epoch= 690: Train set: Average loss: 0.2524, Accuracy: 1176/1284 (91.59%)\n",
      "Valid set: Average loss: 0.2493, Accuracy: 288/320 (90.00%)\n",
      "epoch= 700: Train set: Average loss: 0.2056, Accuracy: 1179/1284 (91.82%)\n",
      "Valid set: Average loss: 0.2612, Accuracy: 288/320 (90.00%)\n",
      "epoch= 710: Train set: Average loss: 0.1989, Accuracy: 1181/1284 (91.98%)\n",
      "Valid set: Average loss: 0.3191, Accuracy: 277/320 (86.56%)\n",
      "epoch= 720: Train set: Average loss: 0.2224, Accuracy: 1167/1284 (90.89%)\n",
      "Valid set: Average loss: 0.2685, Accuracy: 284/320 (88.75%)\n",
      "epoch= 730: Train set: Average loss: 0.4036, Accuracy: 1178/1284 (91.74%)\n",
      "Valid set: Average loss: 0.5819, Accuracy: 252/320 (78.75%)\n",
      "epoch= 740: Train set: Average loss: 0.2741, Accuracy: 1169/1284 (91.04%)\n",
      "Valid set: Average loss: 0.3960, Accuracy: 260/320 (81.25%)\n",
      "epoch= 750: Train set: Average loss: 0.2978, Accuracy: 1168/1284 (90.97%)\n",
      "Valid set: Average loss: 0.2595, Accuracy: 284/320 (88.75%)\n",
      "epoch= 760: Train set: Average loss: 0.2331, Accuracy: 1145/1284 (89.17%)\n",
      "Valid set: Average loss: 0.2418, Accuracy: 293/320 (91.56%)\n",
      "epoch= 770: Train set: Average loss: 0.3424, Accuracy: 1150/1284 (89.56%)\n",
      "Valid set: Average loss: 0.2897, Accuracy: 283/320 (88.44%)\n",
      "epoch= 780: Train set: Average loss: 0.2199, Accuracy: 1158/1284 (90.19%)\n",
      "Valid set: Average loss: 0.2244, Accuracy: 291/320 (90.94%)\n",
      "best_loss updated:  0.22437512874603271\n",
      "epoch= 790: Train set: Average loss: 0.2476, Accuracy: 1169/1284 (91.04%)\n",
      "Valid set: Average loss: 0.2321, Accuracy: 291/320 (90.94%)\n",
      "epoch= 800: Train set: Average loss: 0.2199, Accuracy: 1164/1284 (90.65%)\n",
      "Valid set: Average loss: 0.2463, Accuracy: 284/320 (88.75%)\n",
      "epoch= 810: Train set: Average loss: 0.2432, Accuracy: 1166/1284 (90.81%)\n",
      "Valid set: Average loss: 0.2536, Accuracy: 284/320 (88.75%)\n",
      "epoch= 820: Train set: Average loss: 0.2576, Accuracy: 1168/1284 (90.97%)\n",
      "Valid set: Average loss: 0.4801, Accuracy: 248/320 (77.50%)\n",
      "epoch= 830: Train set: Average loss: 0.1939, Accuracy: 1182/1284 (92.06%)\n",
      "Valid set: Average loss: 0.3204, Accuracy: 281/320 (87.81%)\n",
      "epoch= 840: Train set: Average loss: 0.2512, Accuracy: 1170/1284 (91.12%)\n",
      "Valid set: Average loss: 0.7994, Accuracy: 267/320 (83.44%)\n",
      "epoch= 850: Train set: Average loss: 0.3181, Accuracy: 1111/1284 (86.53%)\n",
      "Valid set: Average loss: 0.2774, Accuracy: 282/320 (88.12%)\n",
      "epoch= 860: Train set: Average loss: 0.2473, Accuracy: 1169/1284 (91.04%)\n",
      "Valid set: Average loss: 0.3989, Accuracy: 280/320 (87.50%)\n",
      "epoch= 870: Train set: Average loss: 0.2566, Accuracy: 1177/1284 (91.67%)\n",
      "Valid set: Average loss: 0.3402, Accuracy: 284/320 (88.75%)\n",
      "epoch= 880: Train set: Average loss: 0.3101, Accuracy: 1179/1284 (91.82%)\n",
      "Valid set: Average loss: 1.0797, Accuracy: 224/320 (70.00%)\n",
      "epoch= 890: Train set: Average loss: 0.2397, Accuracy: 1169/1284 (91.04%)\n",
      "Valid set: Average loss: 0.2657, Accuracy: 291/320 (90.94%)\n",
      "epoch= 900: Train set: Average loss: 0.1926, Accuracy: 1176/1284 (91.59%)\n",
      "Valid set: Average loss: 0.3640, Accuracy: 278/320 (86.88%)\n",
      "epoch= 910: Train set: Average loss: 0.2594, Accuracy: 1143/1284 (89.02%)\n",
      "Valid set: Average loss: 1.2960, Accuracy: 219/320 (68.44%)\n",
      "epoch= 920: Train set: Average loss: 0.3522, Accuracy: 1151/1284 (89.64%)\n",
      "Valid set: Average loss: 0.2783, Accuracy: 281/320 (87.81%)\n",
      "epoch= 930: Train set: Average loss: 0.2028, Accuracy: 1184/1284 (92.21%)\n",
      "Valid set: Average loss: 0.2559, Accuracy: 291/320 (90.94%)\n",
      "epoch= 940: Train set: Average loss: 0.2712, Accuracy: 1174/1284 (91.43%)\n",
      "Valid set: Average loss: 0.2519, Accuracy: 292/320 (91.25%)\n",
      "epoch= 950: Train set: Average loss: 0.1893, Accuracy: 1189/1284 (92.60%)\n",
      "Valid set: Average loss: 0.2312, Accuracy: 290/320 (90.62%)\n",
      "epoch= 960: Train set: Average loss: 0.2720, Accuracy: 1171/1284 (91.20%)\n",
      "Valid set: Average loss: 0.2808, Accuracy: 286/320 (89.38%)\n",
      "epoch= 970: Train set: Average loss: 0.2290, Accuracy: 1176/1284 (91.59%)\n",
      "Valid set: Average loss: 0.3239, Accuracy: 278/320 (86.88%)\n",
      "epoch= 980: Train set: Average loss: 0.2354, Accuracy: 1181/1284 (91.98%)\n",
      "Valid set: Average loss: 0.2713, Accuracy: 290/320 (90.62%)\n",
      "epoch= 990: Train set: Average loss: 0.1984, Accuracy: 1183/1284 (92.13%)\n",
      "Valid set: Average loss: 0.2354, Accuracy: 289/320 (90.31%)\n",
      "epoch= 1000: Train set: Average loss: 0.2219, Accuracy: 1180/1284 (91.90%)\n",
      "Valid set: Average loss: 0.2837, Accuracy: 284/320 (88.75%)\n",
      "epoch= 1010: Train set: Average loss: 0.2102, Accuracy: 1176/1284 (91.59%)\n",
      "Valid set: Average loss: 0.2348, Accuracy: 288/320 (90.00%)\n",
      "epoch= 1020: Train set: Average loss: 0.2309, Accuracy: 1190/1284 (92.68%)\n",
      "Valid set: Average loss: 0.2298, Accuracy: 286/320 (89.38%)\n",
      "epoch= 1030: Train set: Average loss: 0.1931, Accuracy: 1185/1284 (92.29%)\n",
      "Valid set: Average loss: 0.2337, Accuracy: 288/320 (90.00%)\n",
      "epoch= 1040: Train set: Average loss: 0.1881, Accuracy: 1197/1284 (93.22%)\n",
      "Valid set: Average loss: 0.2239, Accuracy: 288/320 (90.00%)\n",
      "best_loss updated:  0.22394330302874246\n",
      "epoch= 1050: Train set: Average loss: 0.1995, Accuracy: 1183/1284 (92.13%)\n",
      "Valid set: Average loss: 0.2282, Accuracy: 287/320 (89.69%)\n",
      "epoch= 1060: Train set: Average loss: 0.1933, Accuracy: 1188/1284 (92.52%)\n",
      "Valid set: Average loss: 0.2341, Accuracy: 289/320 (90.31%)\n",
      "epoch= 1070: Train set: Average loss: 0.3151, Accuracy: 1187/1284 (92.45%)\n",
      "Valid set: Average loss: 0.2284, Accuracy: 288/320 (90.00%)\n",
      "epoch= 1080: Train set: Average loss: 0.2148, Accuracy: 1179/1284 (91.82%)\n",
      "Valid set: Average loss: 0.2376, Accuracy: 290/320 (90.62%)\n",
      "epoch= 1090: Train set: Average loss: 0.1802, Accuracy: 1193/1284 (92.91%)\n",
      "Valid set: Average loss: 0.2356, Accuracy: 288/320 (90.00%)\n",
      "epoch= 1100: Train set: Average loss: 0.1726, Accuracy: 1193/1284 (92.91%)\n",
      "Valid set: Average loss: 0.2341, Accuracy: 288/320 (90.00%)\n",
      "epoch= 1110: Train set: Average loss: 0.2570, Accuracy: 1198/1284 (93.30%)\n",
      "Valid set: Average loss: 0.2428, Accuracy: 288/320 (90.00%)\n",
      "epoch= 1120: Train set: Average loss: 0.2051, Accuracy: 1196/1284 (93.15%)\n",
      "Valid set: Average loss: 0.2253, Accuracy: 290/320 (90.62%)\n",
      "epoch= 1130: Train set: Average loss: 0.1994, Accuracy: 1172/1284 (91.28%)\n",
      "Valid set: Average loss: 0.2239, Accuracy: 289/320 (90.31%)\n",
      "best_loss updated:  0.22388065854708353\n",
      "epoch= 1140: Train set: Average loss: 0.2705, Accuracy: 1183/1284 (92.13%)\n",
      "Valid set: Average loss: 0.2381, Accuracy: 289/320 (90.31%)\n",
      "epoch= 1150: Train set: Average loss: 0.1978, Accuracy: 1196/1284 (93.15%)\n",
      "Valid set: Average loss: 0.2293, Accuracy: 292/320 (91.25%)\n",
      "epoch= 1160: Train set: Average loss: 0.1951, Accuracy: 1187/1284 (92.45%)\n",
      "Valid set: Average loss: 0.2348, Accuracy: 290/320 (90.62%)\n",
      "epoch= 1170: Train set: Average loss: 0.2823, Accuracy: 1196/1284 (93.15%)\n",
      "Valid set: Average loss: 0.2362, Accuracy: 287/320 (89.69%)\n",
      "epoch= 1180: Train set: Average loss: 0.2775, Accuracy: 1185/1284 (92.29%)\n",
      "Valid set: Average loss: 0.2320, Accuracy: 289/320 (90.31%)\n",
      "epoch= 1190: Train set: Average loss: 0.1934, Accuracy: 1199/1284 (93.38%)\n",
      "Valid set: Average loss: 0.2364, Accuracy: 287/320 (89.69%)\n",
      "epoch= 1200: Train set: Average loss: 0.1666, Accuracy: 1202/1284 (93.61%)\n",
      "Valid set: Average loss: 0.2384, Accuracy: 291/320 (90.94%)\n",
      "epoch= 1210: Train set: Average loss: 0.3614, Accuracy: 1187/1284 (92.45%)\n",
      "Valid set: Average loss: 0.2382, Accuracy: 289/320 (90.31%)\n",
      "epoch= 1220: Train set: Average loss: 0.1727, Accuracy: 1205/1284 (93.85%)\n",
      "Valid set: Average loss: 0.2358, Accuracy: 289/320 (90.31%)\n",
      "epoch= 1230: Train set: Average loss: 0.1733, Accuracy: 1199/1284 (93.38%)\n",
      "Valid set: Average loss: 0.2346, Accuracy: 288/320 (90.00%)\n",
      "epoch= 1240: Train set: Average loss: 0.1873, Accuracy: 1200/1284 (93.46%)\n",
      "Valid set: Average loss: 0.2331, Accuracy: 290/320 (90.62%)\n",
      "epoch= 1250: Train set: Average loss: 0.1950, Accuracy: 1188/1284 (92.52%)\n",
      "Valid set: Average loss: 0.2332, Accuracy: 289/320 (90.31%)\n",
      "epoch= 1260: Train set: Average loss: 0.1815, Accuracy: 1197/1284 (93.22%)\n",
      "Valid set: Average loss: 0.2330, Accuracy: 289/320 (90.31%)\n",
      "epoch= 1270: Train set: Average loss: 0.1724, Accuracy: 1202/1284 (93.61%)\n",
      "Valid set: Average loss: 0.2321, Accuracy: 290/320 (90.62%)\n",
      "epoch= 1280: Train set: Average loss: 0.1874, Accuracy: 1195/1284 (93.07%)\n",
      "Valid set: Average loss: 0.2434, Accuracy: 287/320 (89.69%)\n",
      "epoch= 1290: Train set: Average loss: 0.1554, Accuracy: 1206/1284 (93.93%)\n",
      "Valid set: Average loss: 0.2359, Accuracy: 290/320 (90.62%)\n",
      "epoch= 1300: Train set: Average loss: 0.1597, Accuracy: 1200/1284 (93.46%)\n",
      "Valid set: Average loss: 0.2351, Accuracy: 290/320 (90.62%)\n"
     ]
    }
   ],
   "source": [
    "kf = KFold(n_splits=fold_num, shuffle=True, random_state=519)\n",
    "counter  = 1\n",
    "mean_err = 0\n",
    "\n",
    "for train_index, valid_index in kf.split(train_df):\n",
    "     \n",
    "    \n",
    "    is_transfer_learning = False\n",
    "    model = lessFilterResNet50()\n",
    "    model_name = 'Trained_model/less_filter_resent50_' + str(counter) + '.db'\n",
    "    # testing snap shot for essemble\n",
    "    optimizer = torch.optim.Adam(model.parameters(), lr=0.005, weight_decay=0.001)\n",
    "\n",
    "    tr = Trainer(\n",
    "            model,\n",
    "            optimizer,\n",
    "            epochs=1300,\n",
    "            milestones=[300, 1000, 1200],\n",
    "            gamma=0.2,\n",
    "            batch_size=128, \n",
    "            use_cuda=True, \n",
    "            gpu_idx=0,\n",
    "            best_model_name=model_name\n",
    "            )\n",
    "    \n",
    "    train = train_df.loc[train_index].reset_index(drop=True)\n",
    "    valid = train_df.loc[valid_index].reset_index(drop=True)\n",
    "\n",
    "    data_augmentation_args = {\n",
    "        'mirror': False, # not useful here\n",
    "        'rotate': True,\n",
    "        'scale': True,\n",
    "        'translation': True,\n",
    "        #'color jitter' : True,\n",
    "    }\n",
    "    tr.train(train, valid_df=valid, is_transfer_learning=is_transfer_learning, data_augmentation_args=data_augmentation_args)\n",
    "\n",
    "    counter += 1\n",
    "    del model, tr, optimizer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Five fold records\n",
    "* losses: 0.229, 0.198, 0.208, 0.22, 0.224\n",
    "* avg loss: 0.2158\n",
    "* accuracies: 90.65%, 91.90%, 91.59%, 89.10%, 90.31%\n",
    "* avg acc: 90.71%\n",
    "\n",
    "Public LB: 0.1685"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n",
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000002EF430F7EF0>, <torchvision.transforms.transforms.Lambda object at 0x000002EF42703080>, <torchvision.transforms.transforms.ToTensor object at 0x000002EF42703898>]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 10: Train set: Average loss: 0.5972, Accuracy: 1017/1604 (63.40%)\n",
      "epoch= 20: Train set: Average loss: 0.4954, Accuracy: 1204/1604 (75.06%)\n",
      "epoch= 30: Train set: Average loss: 0.4203, Accuracy: 1293/1604 (80.61%)\n",
      "epoch= 40: Train set: Average loss: 0.3621, Accuracy: 1317/1604 (82.11%)\n",
      "epoch= 50: Train set: Average loss: 0.3296, Accuracy: 1357/1604 (84.60%)\n",
      "epoch= 60: Train set: Average loss: 0.2887, Accuracy: 1380/1604 (86.03%)\n",
      "epoch= 70: Train set: Average loss: 0.2950, Accuracy: 1403/1604 (87.47%)\n",
      "epoch= 80: Train set: Average loss: 0.2810, Accuracy: 1396/1604 (87.03%)\n",
      "epoch= 90: Train set: Average loss: 0.2836, Accuracy: 1408/1604 (87.78%)\n",
      "epoch= 100: Train set: Average loss: 0.2819, Accuracy: 1414/1604 (88.15%)\n",
      "epoch= 110: Train set: Average loss: 0.2850, Accuracy: 1408/1604 (87.78%)\n",
      "epoch= 120: Train set: Average loss: 0.2570, Accuracy: 1435/1604 (89.46%)\n",
      "epoch= 130: Train set: Average loss: 0.2449, Accuracy: 1434/1604 (89.40%)\n",
      "epoch= 140: Train set: Average loss: 0.2487, Accuracy: 1441/1604 (89.84%)\n",
      "epoch= 150: Train set: Average loss: 0.2495, Accuracy: 1436/1604 (89.53%)\n",
      "epoch= 160: Train set: Average loss: 0.2431, Accuracy: 1446/1604 (90.15%)\n",
      "epoch= 170: Train set: Average loss: 0.2421, Accuracy: 1450/1604 (90.40%)\n",
      "epoch= 180: Train set: Average loss: 0.2422, Accuracy: 1435/1604 (89.46%)\n",
      "epoch= 190: Train set: Average loss: 0.2687, Accuracy: 1428/1604 (89.03%)\n",
      "epoch= 200: Train set: Average loss: 0.2446, Accuracy: 1433/1604 (89.34%)\n",
      "epoch= 210: Train set: Average loss: 0.2542, Accuracy: 1434/1604 (89.40%)\n",
      "epoch= 220: Train set: Average loss: 0.2389, Accuracy: 1441/1604 (89.84%)\n",
      "epoch= 230: Train set: Average loss: 0.2472, Accuracy: 1430/1604 (89.15%)\n",
      "epoch= 240: Train set: Average loss: 0.2575, Accuracy: 1442/1604 (89.90%)\n",
      "epoch= 250: Train set: Average loss: 0.2516, Accuracy: 1444/1604 (90.02%)\n",
      "epoch= 260: Train set: Average loss: 0.2524, Accuracy: 1430/1604 (89.15%)\n",
      "epoch= 270: Train set: Average loss: 0.2387, Accuracy: 1443/1604 (89.96%)\n",
      "epoch= 280: Train set: Average loss: 0.2518, Accuracy: 1434/1604 (89.40%)\n",
      "epoch= 290: Train set: Average loss: 0.2339, Accuracy: 1450/1604 (90.40%)\n",
      "epoch= 300: Train set: Average loss: 0.2373, Accuracy: 1444/1604 (90.02%)\n",
      "epoch= 310: Train set: Average loss: 0.1987, Accuracy: 1469/1604 (91.58%)\n",
      "epoch= 320: Train set: Average loss: 0.1966, Accuracy: 1485/1604 (92.58%)\n",
      "epoch= 330: Train set: Average loss: 0.1934, Accuracy: 1477/1604 (92.08%)\n",
      "epoch= 340: Train set: Average loss: 0.1800, Accuracy: 1491/1604 (92.96%)\n",
      "epoch= 350: Train set: Average loss: 0.1843, Accuracy: 1479/1604 (92.21%)\n",
      "epoch= 360: Train set: Average loss: 0.1797, Accuracy: 1482/1604 (92.39%)\n",
      "epoch= 370: Train set: Average loss: 0.1782, Accuracy: 1488/1604 (92.77%)\n",
      "epoch= 380: Train set: Average loss: 0.1726, Accuracy: 1483/1604 (92.46%)\n",
      "epoch= 390: Train set: Average loss: 0.1749, Accuracy: 1502/1604 (93.64%)\n",
      "epoch= 400: Train set: Average loss: 0.1891, Accuracy: 1480/1604 (92.27%)\n",
      "epoch= 410: Train set: Average loss: 0.1708, Accuracy: 1496/1604 (93.27%)\n",
      "epoch= 420: Train set: Average loss: 0.1577, Accuracy: 1504/1604 (93.77%)\n",
      "epoch= 430: Train set: Average loss: 0.1804, Accuracy: 1493/1604 (93.08%)\n",
      "epoch= 440: Train set: Average loss: 0.1758, Accuracy: 1486/1604 (92.64%)\n",
      "epoch= 450: Train set: Average loss: 0.1616, Accuracy: 1504/1604 (93.77%)\n",
      "epoch= 460: Train set: Average loss: 0.1580, Accuracy: 1499/1604 (93.45%)\n",
      "epoch= 470: Train set: Average loss: 0.1780, Accuracy: 1487/1604 (92.71%)\n",
      "epoch= 480: Train set: Average loss: 0.1600, Accuracy: 1490/1604 (92.89%)\n",
      "epoch= 490: Train set: Average loss: 0.1631, Accuracy: 1508/1604 (94.01%)\n",
      "epoch= 500: Train set: Average loss: 0.1592, Accuracy: 1510/1604 (94.14%)\n",
      "epoch= 510: Train set: Average loss: 0.1525, Accuracy: 1506/1604 (93.89%)\n",
      "epoch= 520: Train set: Average loss: 0.1444, Accuracy: 1513/1604 (94.33%)\n",
      "epoch= 530: Train set: Average loss: 0.1765, Accuracy: 1486/1604 (92.64%)\n",
      "epoch= 540: Train set: Average loss: 0.1602, Accuracy: 1504/1604 (93.77%)\n",
      "epoch= 550: Train set: Average loss: 0.1604, Accuracy: 1496/1604 (93.27%)\n",
      "epoch= 560: Train set: Average loss: 0.1630, Accuracy: 1503/1604 (93.70%)\n",
      "epoch= 570: Train set: Average loss: 0.1554, Accuracy: 1507/1604 (93.95%)\n",
      "epoch= 580: Train set: Average loss: 0.1548, Accuracy: 1500/1604 (93.52%)\n",
      "epoch= 590: Train set: Average loss: 0.1611, Accuracy: 1507/1604 (93.95%)\n",
      "epoch= 600: Train set: Average loss: 0.1721, Accuracy: 1486/1604 (92.64%)\n",
      "epoch= 610: Train set: Average loss: 0.1423, Accuracy: 1517/1604 (94.58%)\n",
      "epoch= 620: Train set: Average loss: 0.1478, Accuracy: 1516/1604 (94.51%)\n",
      "epoch= 630: Train set: Average loss: 0.1395, Accuracy: 1515/1604 (94.45%)\n",
      "epoch= 640: Train set: Average loss: 0.1513, Accuracy: 1518/1604 (94.64%)\n",
      "epoch= 650: Train set: Average loss: 0.1367, Accuracy: 1520/1604 (94.76%)\n",
      "epoch= 660: Train set: Average loss: 0.1642, Accuracy: 1501/1604 (93.58%)\n",
      "epoch= 670: Train set: Average loss: 0.1492, Accuracy: 1512/1604 (94.26%)\n",
      "epoch= 680: Train set: Average loss: 0.1314, Accuracy: 1519/1604 (94.70%)\n",
      "epoch= 690: Train set: Average loss: 0.1370, Accuracy: 1513/1604 (94.33%)\n",
      "epoch= 700: Train set: Average loss: 0.1396, Accuracy: 1512/1604 (94.26%)\n",
      "epoch= 710: Train set: Average loss: 0.1312, Accuracy: 1528/1604 (95.26%)\n",
      "epoch= 720: Train set: Average loss: 0.1364, Accuracy: 1526/1604 (95.14%)\n",
      "epoch= 730: Train set: Average loss: 0.1363, Accuracy: 1519/1604 (94.70%)\n",
      "epoch= 740: Train set: Average loss: 0.1504, Accuracy: 1518/1604 (94.64%)\n",
      "epoch= 750: Train set: Average loss: 0.1312, Accuracy: 1518/1604 (94.64%)\n",
      "epoch= 760: Train set: Average loss: 0.1203, Accuracy: 1521/1604 (94.83%)\n",
      "epoch= 770: Train set: Average loss: 0.1496, Accuracy: 1514/1604 (94.39%)\n",
      "epoch= 780: Train set: Average loss: 0.1200, Accuracy: 1535/1604 (95.70%)\n",
      "epoch= 790: Train set: Average loss: 0.1340, Accuracy: 1524/1604 (95.01%)\n",
      "epoch= 800: Train set: Average loss: 0.1420, Accuracy: 1523/1604 (94.95%)\n",
      "epoch= 810: Train set: Average loss: 0.1194, Accuracy: 1530/1604 (95.39%)\n",
      "epoch= 820: Train set: Average loss: 0.1342, Accuracy: 1522/1604 (94.89%)\n",
      "epoch= 830: Train set: Average loss: 0.1272, Accuracy: 1518/1604 (94.64%)\n",
      "epoch= 840: Train set: Average loss: 0.1209, Accuracy: 1528/1604 (95.26%)\n",
      "epoch= 850: Train set: Average loss: 0.1355, Accuracy: 1531/1604 (95.45%)\n",
      "epoch= 860: Train set: Average loss: 0.1413, Accuracy: 1520/1604 (94.76%)\n",
      "epoch= 870: Train set: Average loss: 0.1171, Accuracy: 1539/1604 (95.95%)\n",
      "epoch= 880: Train set: Average loss: 0.1426, Accuracy: 1526/1604 (95.14%)\n",
      "epoch= 890: Train set: Average loss: 0.1173, Accuracy: 1541/1604 (96.07%)\n",
      "epoch= 900: Train set: Average loss: 0.1270, Accuracy: 1530/1604 (95.39%)\n",
      "epoch= 910: Train set: Average loss: 0.1195, Accuracy: 1537/1604 (95.82%)\n",
      "epoch= 920: Train set: Average loss: 0.1327, Accuracy: 1526/1604 (95.14%)\n",
      "epoch= 930: Train set: Average loss: 0.1324, Accuracy: 1521/1604 (94.83%)\n",
      "epoch= 940: Train set: Average loss: 0.1387, Accuracy: 1524/1604 (95.01%)\n",
      "epoch= 950: Train set: Average loss: 0.1166, Accuracy: 1532/1604 (95.51%)\n",
      "epoch= 960: Train set: Average loss: 0.1445, Accuracy: 1515/1604 (94.45%)\n",
      "epoch= 970: Train set: Average loss: 0.1332, Accuracy: 1525/1604 (95.07%)\n",
      "epoch= 980: Train set: Average loss: 0.1093, Accuracy: 1533/1604 (95.57%)\n",
      "epoch= 990: Train set: Average loss: 0.1179, Accuracy: 1533/1604 (95.57%)\n",
      "epoch= 1000: Train set: Average loss: 0.1259, Accuracy: 1524/1604 (95.01%)\n",
      "epoch= 1010: Train set: Average loss: 0.0914, Accuracy: 1550/1604 (96.63%)\n",
      "epoch= 1020: Train set: Average loss: 0.0936, Accuracy: 1548/1604 (96.51%)\n",
      "epoch= 1030: Train set: Average loss: 0.0961, Accuracy: 1547/1604 (96.45%)\n",
      "epoch= 1040: Train set: Average loss: 0.0819, Accuracy: 1557/1604 (97.07%)\n",
      "epoch= 1050: Train set: Average loss: 0.0810, Accuracy: 1558/1604 (97.13%)\n",
      "epoch= 1060: Train set: Average loss: 0.0768, Accuracy: 1556/1604 (97.01%)\n",
      "epoch= 1070: Train set: Average loss: 0.0699, Accuracy: 1569/1604 (97.82%)\n",
      "epoch= 1080: Train set: Average loss: 0.0835, Accuracy: 1553/1604 (96.82%)\n",
      "epoch= 1090: Train set: Average loss: 0.0698, Accuracy: 1560/1604 (97.26%)\n",
      "epoch= 1100: Train set: Average loss: 0.0697, Accuracy: 1570/1604 (97.88%)\n",
      "epoch= 1110: Train set: Average loss: 0.0872, Accuracy: 1554/1604 (96.88%)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch= 1120: Train set: Average loss: 0.0747, Accuracy: 1556/1604 (97.01%)\n",
      "epoch= 1130: Train set: Average loss: 0.0673, Accuracy: 1563/1604 (97.44%)\n",
      "epoch= 1140: Train set: Average loss: 0.0761, Accuracy: 1561/1604 (97.32%)\n",
      "epoch= 1150: Train set: Average loss: 0.0710, Accuracy: 1564/1604 (97.51%)\n",
      "epoch= 1160: Train set: Average loss: 0.0726, Accuracy: 1559/1604 (97.19%)\n",
      "epoch= 1170: Train set: Average loss: 0.0697, Accuracy: 1558/1604 (97.13%)\n",
      "epoch= 1180: Train set: Average loss: 0.0612, Accuracy: 1571/1604 (97.94%)\n",
      "epoch= 1190: Train set: Average loss: 0.0765, Accuracy: 1558/1604 (97.13%)\n",
      "epoch= 1200: Train set: Average loss: 0.0664, Accuracy: 1561/1604 (97.32%)\n",
      "epoch= 1210: Train set: Average loss: 0.0719, Accuracy: 1564/1604 (97.51%)\n",
      "epoch= 1220: Train set: Average loss: 0.0538, Accuracy: 1566/1604 (97.63%)\n",
      "epoch= 1230: Train set: Average loss: 0.0540, Accuracy: 1570/1604 (97.88%)\n",
      "epoch= 1240: Train set: Average loss: 0.0586, Accuracy: 1568/1604 (97.76%)\n",
      "epoch= 1250: Train set: Average loss: 0.0534, Accuracy: 1578/1604 (98.38%)\n",
      "epoch= 1260: Train set: Average loss: 0.0573, Accuracy: 1566/1604 (97.63%)\n",
      "epoch= 1270: Train set: Average loss: 0.0650, Accuracy: 1569/1604 (97.82%)\n",
      "epoch= 1280: Train set: Average loss: 0.0501, Accuracy: 1574/1604 (98.13%)\n",
      "epoch= 1290: Train set: Average loss: 0.0600, Accuracy: 1567/1604 (97.69%)\n",
      "epoch= 1300: Train set: Average loss: 0.0678, Accuracy: 1566/1604 (97.63%)\n"
     ]
    }
   ],
   "source": [
    "model = lessFilterResNet50()\n",
    "model_name = 'Trained_model/less_filter_resent50_all.db'\n",
    "    \n",
    "# testing snap shot for essemble\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=0.005, weight_decay=0.001)\n",
    "\n",
    "tr = Trainer(\n",
    "        model,\n",
    "        optimizer,\n",
    "        epochs=1300,\n",
    "        milestones=[300, 1000, 1200],\n",
    "        gamma=0.2,\n",
    "        batch_size=128, \n",
    "        use_cuda=True, \n",
    "        gpu_idx=0,\n",
    "        best_model_name=model_name)\n",
    "\n",
    "data_augmentation_args = {\n",
    "    'mirror': False, # not useful here\n",
    "    'rotate': True,\n",
    "    'scale': True,\n",
    "    'translation': True,\n",
    "    #'color jitter' : True,\n",
    "}\n",
    "tr.train(train_df, is_transfer_learning=is_transfer_learning, data_augmentation_args=data_augmentation_args)\n",
    "tr.save(model_name)\n",
    "del model, optimizer, tr"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Predict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(8424, 4)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>band_1</th>\n",
       "      <th>band_2</th>\n",
       "      <th>id</th>\n",
       "      <th>inc_angle</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[-15.863251, -15.201077, -17.887735, -19.17248...</td>\n",
       "      <td>[-21.629612, -21.142353, -23.908337, -28.34524...</td>\n",
       "      <td>5941774d</td>\n",
       "      <td>34.966400</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[-26.058969497680664, -26.058969497680664, -26...</td>\n",
       "      <td>[-25.754207611083984, -25.754207611083984, -25...</td>\n",
       "      <td>4023181e</td>\n",
       "      <td>32.615072</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[-14.14109992980957, -15.064241409301758, -17....</td>\n",
       "      <td>[-14.74563980102539, -14.590410232543945, -14....</td>\n",
       "      <td>b20200e4</td>\n",
       "      <td>37.505433</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[-12.167478, -13.706167, -16.54837, -13.572674...</td>\n",
       "      <td>[-24.32222, -26.375538, -24.096739, -23.8769, ...</td>\n",
       "      <td>e7f018bb</td>\n",
       "      <td>34.473900</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[-23.37459373474121, -26.02718162536621, -28.1...</td>\n",
       "      <td>[-25.72234344482422, -27.011577606201172, -23....</td>\n",
       "      <td>4371c8c3</td>\n",
       "      <td>43.918874</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                              band_1  \\\n",
       "0  [-15.863251, -15.201077, -17.887735, -19.17248...   \n",
       "1  [-26.058969497680664, -26.058969497680664, -26...   \n",
       "2  [-14.14109992980957, -15.064241409301758, -17....   \n",
       "3  [-12.167478, -13.706167, -16.54837, -13.572674...   \n",
       "4  [-23.37459373474121, -26.02718162536621, -28.1...   \n",
       "\n",
       "                                              band_2        id  inc_angle  \n",
       "0  [-21.629612, -21.142353, -23.908337, -28.34524...  5941774d  34.966400  \n",
       "1  [-25.754207611083984, -25.754207611083984, -25...  4023181e  32.615072  \n",
       "2  [-14.74563980102539, -14.590410232543945, -14....  b20200e4  37.505433  \n",
       "3  [-24.32222, -26.375538, -24.096739, -23.8769, ...  e7f018bb  34.473900  \n",
       "4  [-25.72234344482422, -27.011577606201172, -23....  4371c8c3  43.918874  "
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df = pd.read_json('Data/test.json')\n",
    "print(test_df.shape)\n",
    "test_df.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0       5941774d\n",
      "1       4023181e\n",
      "2       b20200e4\n",
      "3       e7f018bb\n",
      "4       4371c8c3\n",
      "5       a8d9b1fd\n",
      "6       29e7727e\n",
      "7       92a51ffb\n",
      "8       c769ac97\n",
      "9       aee0547d\n",
      "10      565b28ac\n",
      "11      e04e9775\n",
      "12      8e8161d1\n",
      "13      4cf4d256\n",
      "14      139e5324\n",
      "15      f156976f\n",
      "16      68a117cc\n",
      "17      d9aa7a56\n",
      "18      9005b143\n",
      "19      5f6d3988\n",
      "20      9ad70954\n",
      "21      b9087b9e\n",
      "22      a39a1427\n",
      "23      82fbe8ed\n",
      "24      1fae4879\n",
      "25      6dd8f13d\n",
      "26      bbad5958\n",
      "27      54527583\n",
      "28      be8fa29c\n",
      "29      81a3328f\n",
      "          ...   \n",
      "8394    8ae30ce6\n",
      "8395    de27ed88\n",
      "8396    66d5196f\n",
      "8397    d85f1858\n",
      "8398    16dcb33a\n",
      "8399    eca3158e\n",
      "8400    08daeee6\n",
      "8401    e9c513ee\n",
      "8402    b1519fa6\n",
      "8403    dfc89540\n",
      "8404    8fd8c0e9\n",
      "8405    45df6347\n",
      "8406    bf7928d7\n",
      "8407    7b587c05\n",
      "8408    c2834388\n",
      "8409    146143c3\n",
      "8410    d59aee00\n",
      "8411    cbc0b93b\n",
      "8412    088e2ff7\n",
      "8413    673d33cd\n",
      "8414    674b031e\n",
      "8415    43db4207\n",
      "8416    156855e1\n",
      "8417    ac96cfb0\n",
      "8418    fe45aef5\n",
      "8419    16ee9b50\n",
      "8420    5a599eb7\n",
      "8421    df30d6dd\n",
      "8422    18af95b1\n",
      "8423    27d788c8\n",
      "Name: id, Length: 8424, dtype: object\n"
     ]
    }
   ],
   "source": [
    "test_ids = test_df['id']\n",
    "print(test_ids)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "del test_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Index(['band_mixed', 'inc_angle', 'is_iceberg'], dtype='object')\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>band_mixed</th>\n",
       "      <th>inc_angle</th>\n",
       "      <th>is_iceberg</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[[[91, 77, 48], [92, 78, 48], [88, 74, 48], [8...</td>\n",
       "      <td>34.966400</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[[[64, 71, 48], [64, 71, 47], [64, 71, 47], [6...</td>\n",
       "      <td>32.615072</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>[[[87, 69, 48], [89, 72, 48], [86, 72, 48], [7...</td>\n",
       "      <td>35.782900</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                           band_mixed  inc_angle  is_iceberg\n",
       "0   [[[91, 77, 48], [92, 78, 48], [88, 74, 48], [8...  34.966400           0\n",
       "1   [[[64, 71, 48], [64, 71, 47], [64, 71, 47], [6...  32.615072           0\n",
       "10  [[[87, 69, 48], [89, 72, 48], [86, 72, 48], [7...  35.782900           0"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df = pd.read_json('Data/combined_test.json')\n",
    "#test_df.reset_index(drop=True, inplace=True)\n",
    "print(test_df.columns)\n",
    "test_df.head(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>band_mixed</th>\n",
       "      <th>inc_angle</th>\n",
       "      <th>is_iceberg</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[[[91, 77, 48], [92, 78, 48], [88, 74, 48], [8...</td>\n",
       "      <td>34.966400</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[[[64, 71, 48], [64, 71, 47], [64, 71, 47], [6...</td>\n",
       "      <td>32.615072</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[[[95, 101, 48], [94, 101, 48], [91, 100, 48],...</td>\n",
       "      <td>37.505433</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[[[98, 74, 48], [96, 72, 48], [92, 74, 48], [9...</td>\n",
       "      <td>34.473900</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[[[67, 73, 48], [64, 71, 48], [62, 75, 48], [6...</td>\n",
       "      <td>43.918874</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                          band_mixed  inc_angle  is_iceberg\n",
       "0  [[[91, 77, 48], [92, 78, 48], [88, 74, 48], [8...  34.966400           0\n",
       "1  [[[64, 71, 48], [64, 71, 47], [64, 71, 47], [6...  32.615072           0\n",
       "2  [[[95, 101, 48], [94, 101, 48], [91, 100, 48],...  37.505433           0\n",
       "3  [[[98, 74, 48], [96, 72, 48], [92, 74, 48], [9...  34.473900           0\n",
       "4  [[[67, 73, 48], [64, 71, 48], [62, 75, 48], [6...  43.918874           0"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df.sort_index(inplace=True)\n",
    "test_df.head(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### check if preprocessed and original test data matches the sequence (prevent forget-to-reset_index error...)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000001E532224048>, <torchvision.transforms.transforms.ToTensor object at 0x000001E532224208>]\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(104, 1)\n"
     ]
    }
   ],
   "source": [
    "is_transfer_learning = False\n",
    "model = lessFilterResNet50()\n",
    "# testing snap shot for essemble\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=0.005, weight_decay=0.001)\n",
    "\n",
    "tr = Trainer(\n",
    "        model,\n",
    "        optimizer,\n",
    "        epochs=1800,\n",
    "        milestones=[300, 1100, 1600],\n",
    "        gamma=0.2,\n",
    "        batch_size=128, \n",
    "        use_cuda=True, \n",
    "        gpu_idx=0)\n",
    "\n",
    "data_augmentation_args = {\n",
    "    'mirror': False, # not useful here\n",
    "    'rotate': True,\n",
    "    'scale': True,\n",
    "    'translation': True,\n",
    "    #'color jitter' : True,\n",
    "}\n",
    "tr.load('Trained_model/less_filter_resent50_all.db')\n",
    "predictions = tr.test(test_df, is_transfer_learning=is_transfer_learning)\n",
    "\n",
    "del model, optimizer, tr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[  2.40571739e-04   5.17799333e-03   7.46687874e-05   9.99246478e-01\n",
      "   8.82943219e-04   2.47403956e-03   1.34727883e-03   9.99104083e-01\n",
      "   3.12092336e-04   3.98847740e-04]\n",
      "(8424,)\n"
     ]
    }
   ],
   "source": [
    "predictions = predictions.reshape((predictions.shape[0],))\n",
    "print(predictions[:10])\n",
    "print(predictions.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "submission = pd.DataFrame()\n",
    "submission['id'] = test_ids\n",
    "submission['is_iceberg'] = predictions\n",
    "submission.to_csv('Submissions/submission_resnet50_all.csv', float_format=\"%.15f\", index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "False    4994\n",
      "True     3430\n",
      "dtype: int64\n"
     ]
    }
   ],
   "source": [
    "print(pd.Series(predictions > 0.5).value_counts())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000001E3040105F8>, <torchvision.transforms.transforms.ToTensor object at 0x000001E304002E80>]\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(104, 1)\n",
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000001E53220A0F0>, <torchvision.transforms.transforms.ToTensor object at 0x000001E53220A438>]\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(104, 1)\n",
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000001E304019BE0>, <torchvision.transforms.transforms.ToTensor object at 0x000001E304019080>]\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(104, 1)\n",
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000001E304023978>, <torchvision.transforms.transforms.ToTensor object at 0x000001E304023400>]\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(104, 1)\n",
      "gpu: 0  available: True\n",
      "Model in use:\n",
      "LessFilterResNet (\n",
      "  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n",
      "  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (relu): ReLU (inplace)\n",
      "  (maxpool): MaxPool2d (size=(3, 3), stride=(2, 2), dilation=(1, 1))\n",
      "  (layer1): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer2): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer3): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (3): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (4): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (5): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (layer4): Sequential (\n",
      "    (0): DropBottleneck (\n",
      "      (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (downsample): Sequential (\n",
      "        (0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)\n",
      "        (1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      )\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (1): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "    (2): DropBottleneck (\n",
      "      (conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n",
      "      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "      (bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True)\n",
      "      (relu): ReLU (inplace)\n",
      "      (drop): Dropout2d (p=0.2)\n",
      "      (drop2): Dropout2d (p=0.2)\n",
      "    )\n",
      "  )\n",
      "  (avgpool): AvgPool2d (size=5, stride=3, padding=0, ceil_mode=True, count_include_pad=True)\n",
      "  (bn2): BatchNorm1d(1, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (fc): Linear (2049 -> 1)\n",
      "  (bn3): BatchNorm1d(256, eps=1e-05, momentum=0.1, affine=True)\n",
      "  (drop): Dropout2d (p=0.1)\n",
      "  (fc2): Linear (256 -> 1)\n",
      "  (prob): Sigmoid ()\n",
      ")\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using:\n",
      " [<torchvision.transforms.transforms.ToPILImage object at 0x000001E30401AEF0>, <torchvision.transforms.transforms.ToTensor object at 0x000001E30401A198>]\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(128, 1)\n",
      "(104, 1)\n"
     ]
    }
   ],
   "source": [
    "submission = pd.DataFrame()\n",
    "submission['id'] = test_ids\n",
    "submission['is_iceberg'] = np.zeros((len(test_ids),))\n",
    "    \n",
    "for counter in range(fold_num):\n",
    "\n",
    "    is_transfer_learning = False\n",
    "    model = lessFilterResNet50()\n",
    "    model_name = 'Trained_model/less_filter_resent50_' + str(counter+1)+ '.db'\n",
    "    # testing snap shot for essemble\n",
    "    optimizer = torch.optim.Adam(model.parameters(), lr=0.005, weight_decay=0.001)\n",
    "\n",
    "    tr = Trainer(\n",
    "            model,\n",
    "            optimizer,\n",
    "            epochs=1800,\n",
    "            milestones=[300, 1100, 1600],\n",
    "            gamma=0.2,\n",
    "            batch_size=128, \n",
    "            use_cuda=True, \n",
    "            gpu_idx=0,\n",
    "            best_model_name = model_name)\n",
    "    \n",
    "    tr.load_checkpoint()\n",
    "    predictions = tr.test(test_df, is_transfer_learning=is_transfer_learning)\n",
    "    predictions = predictions.reshape((predictions.shape[0],))\n",
    "    submission['is_iceberg'] += predictions\n",
    "    \n",
    "    del model, optimizer, tr\n",
    "    \n",
    "submission['is_iceberg'] /= fold_num * 1.0\n",
    "submission.to_csv('Submissions/submission_resnet50_out_of_fold_avg.csv', float_format=\"%.15f\", index=False)    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "False    4994\n",
      "True     3430\n",
      "Name: is_iceberg, dtype: int64\n",
      "False    6386\n",
      "True     2038\n",
      "Name: is_iceberg, dtype: int64\n",
      "True     6404\n",
      "False    2020\n",
      "Name: is_iceberg, dtype: int64\n"
     ]
    }
   ],
   "source": [
    "sub_all = pd.read_csv('Submissions/submission_resnet50_all.csv')\n",
    "sub_comb = pd.read_csv('Submissions/submission_resnet50_out_of_fold_avg.csv')\n",
    "print(pd.Series(sub_all['is_iceberg'] > 0.5).value_counts())\n",
    "print(pd.Series(sub_comb['is_iceberg'] > 0.5).value_counts())\n",
    "\n",
    "result1 = sub_all['is_iceberg'] > 0.5\n",
    "result2 = sub_comb['is_iceberg'] > 0.5\n",
    "print(pd.Series(result1 == result2).value_counts())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## cross validate fine-tune @ fc layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "kf = KFold(n_splits=fold_num, shuffle=True, random_state=519)\n",
    "counter  = 0\n",
    "mean_err = 0\n",
    "\n",
    "for train_index, valid_index in kf.split(train_df):\n",
    "     \n",
    "    is_transfer_learning = True\n",
    "    model, to_train_dict = fineTuneResNet50(pretrained=True)\n",
    "\n",
    "    print(\"Dict. to train:\")\n",
    "    print(to_train_dict.keys())\n",
    "    \n",
    "    # set only the parameters in to_train_dict to train\n",
    "    # others: requires_grad = False\n",
    "    to_train_parmas = []\n",
    "    for name, param in model.named_parameters():\n",
    "        if name in to_train_dict.keys():\n",
    "            print('\\nTrain params:', name)\n",
    "            to_train_parmas.append(param)\n",
    "        else:\n",
    "            print(name, end=\" \")\n",
    "            param.requires_grad = False\n",
    "    \n",
    "    optimizer = torch.optim.Adam(to_train_parmas, lr=0.01, weight_decay=0.001)\n",
    "    \n",
    "    '''\n",
    "    is_transfer_learning = False\n",
    "    model = lessFilterResNet50()\n",
    "    # testing snap shot for essemble\n",
    "    optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=0.001)\n",
    "    '''\n",
    "    tr = Trainer(\n",
    "            model,\n",
    "            optimizer,\n",
    "            epochs=1000,\n",
    "            milestones=[50, 600],\n",
    "            gamma=0.1,\n",
    "            batch_size=128, \n",
    "            use_cuda=True, \n",
    "            gpu_idx=0)\n",
    "    train = train_df.loc[train_index].reset_index(drop=True)\n",
    "    valid = train_df.loc[valid_index].reset_index(drop=True)\n",
    "\n",
    "    tr.train(train, valid_df=valid, is_transfer_learning=is_transfer_learning)\n",
    "    \n",
    "    counter += 1\n",
    "    tr.save('Trained_model/fine_tune_resent50_' + str(counter) + '.db')\n",
    "    del model, tr, optimizer"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}