Submission Generation.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from trainer import Trainer\n",
    "from meta_featured_trainer import MetaFeatureTrainer\n",
    "from resnet import fineTuneResNet50,lessFilterResNet50, paperResNet18, nnResNet, get_ensemble_resnet\n",
    "from senet import senetXX_generic\n",
    "from swwae_trainer import SWWAETrainer\n",
    "from swwae import get_swwae\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "%matplotlib notebook"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import KFold\n",
    "fold_num = 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "from torch.optim import lr_scheduler\n",
    "import gc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_ids = pd.read_csv('Data/test_ids.csv')['id']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "7"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gc.enable()\n",
    "gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Index(['band_mixed', 'inc_angle', 'is_iceberg'], dtype='object')\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\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>[[[0.3602381271, 0.3185865893, 0.6397618729, 0...</td>\n",
       "      <td>34.966400</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[[[0.23705839, 0.2707230281, 0.76294161, 0.729...</td>\n",
       "      <td>32.615072</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>[[[0.3381528741, 0.2572200913, 0.6618471259, 0...</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   [[[0.3602381271, 0.3185865893, 0.6397618729, 0...  34.966400           0\n",
       "1   [[[0.23705839, 0.2707230281, 0.76294161, 0.729...  32.615072           0\n",
       "10  [[[0.3381528741, 0.2572200913, 0.6618471259, 0...  35.782900           0"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df = pd.read_json('Data/polar_split_no_denoise_processed_test.json', dtype={'inc_angle': str})\n",
    "test_df.inc_angle = pd.to_numeric(test_df.inc_angle, errors='coerce')\n",
    "\n",
    "#test_df.inc_angle = (test_df.inc_angle / 0.3).astype(np.int)\n",
    "\n",
    "#test_df.reset_index(drop=True, inplace=True)\n",
    "print(test_df.columns)\n",
    "test_df.head(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\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>[[[0.3602381271, 0.3185865893, 0.6397618729, 0...</td>\n",
       "      <td>34.966400</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[[[0.23705839, 0.2707230281, 0.76294161, 0.729...</td>\n",
       "      <td>32.615072</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[[[0.381044323, 0.39847112970000004, 0.6189556...</td>\n",
       "      <td>37.505433</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[[[0.4048886681, 0.28734042120000003, 0.595111...</td>\n",
       "      <td>34.473900</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[[[0.269489719, 0.2710927935, 0.730510281, 0.7...</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  [[[0.3602381271, 0.3185865893, 0.6397618729, 0...  34.966400           0\n",
       "1  [[[0.23705839, 0.2707230281, 0.76294161, 0.729...  32.615072           0\n",
       "2  [[[0.381044323, 0.39847112970000004, 0.6189556...  37.505433           0\n",
       "3  [[[0.4048886681, 0.28734042120000003, 0.595111...  34.473900           0\n",
       "4  [[[0.269489719, 0.2710927935, 0.730510281, 0.7...  43.918874           0"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df.sort_index(inplace=True)\n",
    "test_df.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "submission = pd.DataFrame()\n",
    "submission['id'] = test_ids\n",
    "submission['is_iceberg'] = np.zeros((len(test_ids),))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def predict_with_oof_models(seed=0):\n",
    "    global fold_num\n",
    "    global test_ids\n",
    "    \n",
    "    final_predictions = []\n",
    "    for counter in range(fold_num):\n",
    "\n",
    "        is_transfer_learning = False\n",
    "        model = get_ensemble_resnet()\n",
    "        #model = nnResNet(meta_features=1)  #dpn92(num_classes=1) #lessFilterResNet50()\n",
    "        model_name= 'Trained_model/ens_resnet_cpl_' + str(counter+1) + '.db'\n",
    "        # model_name = 'Trained_model/nn_resnet_soft_pl_n_valid_' + 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",
    "                seed=seed,\n",
    "                verbose=0)\n",
    "\n",
    "        tr.load_checkpoint()\n",
    "        \n",
    "        data_augmentation_args = {\n",
    "            'mirror': False, # not useful here\n",
    "            'rotate': True,\n",
    "            'scale': True,\n",
    "            'translation': True\n",
    "        }\n",
    "        \n",
    "        predictions = tr.test(test_df, is_transfer_learning=is_transfer_learning, is_augment=True, data_augmentation_args=data_augmentation_args)\n",
    "        predictions = predictions.reshape((predictions.shape[0],))\n",
    "        final_predictions.append(predictions)\n",
    "\n",
    "        del model, optimizer, tr\n",
    "        \n",
    "    return final_predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "1 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "2 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "3 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "4 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "5 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "6 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "7 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "8 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "9 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "10 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "11 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "12 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "13 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "14 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "15 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "16 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "17 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "18 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "19 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "20 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "21 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "22 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "23 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "24 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "25 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "26 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "27 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "28 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "29 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "30 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "31 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "32 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "33 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "34 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "35 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "36 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "37 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "38 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n",
      "39 round\n",
      "gpu: 0  available: True\n",
      "epoch= 106 best_loss= 0.05816717385314405\n",
      "gpu: 0  available: True\n",
      "epoch= 135 best_loss= 0.07412643822220465\n",
      "gpu: 0  available: True\n",
      "epoch= 98 best_loss= 0.07048591789789498\n",
      "gpu: 0  available: True\n",
      "epoch= 80 best_loss= 0.07901024012826383\n",
      "gpu: 0  available: True\n",
      "epoch= 110 best_loss= 0.0714154920152699\n"
     ]
    }
   ],
   "source": [
    "repeat_count = 40\n",
    "for i in range(repeat_count):\n",
    "    print(i, 'round')\n",
    "    predictions = predict_with_oof_models(seed=i)\n",
    "    for j, p in enumerate(predictions):\n",
    "        submission['is_iceberg_r' + str(i) + '_p' + str(j) ] = p"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "submission['is_iceberg_max'] = submission.iloc[:, 2:repeat_count*fold_num+2].max(axis=1)\n",
    "submission['is_iceberg_min'] = submission.iloc[:, 2:repeat_count*fold_num+2].min(axis=1)\n",
    "submission['is_iceberg_mean'] = submission.iloc[:, 2:repeat_count*fold_num+2].mean(axis=1)\n",
    "submission['is_iceberg_median'] = submission.iloc[:, 2:repeat_count*fold_num+2].median(axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "submission['is_iceberg'] = submission['is_iceberg_mean']\n",
    "submission[['id', 'is_iceberg']].to_csv('Submissions/submission_ensemble_resnet_cpl_oof_augment_avg.csv', \n",
    "                                        index=False, float_format='%.15f')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cutoff_hi=.1\n",
    "cutoff_lo=.9\n",
    "submission['is_iceberg'] = np.where(np.all(submission.iloc[:,2:repeat_count*fold_num+2] > cutoff_lo, axis=1), \n",
    "                                    submission['is_iceberg_max'], # all > lo\n",
    "                                    np.where(np.all(submission.iloc[:,2:repeat_count*fold_num+2] < cutoff_hi, axis=1),\n",
    "                                             submission['is_iceberg_min'], # all < hi \n",
    "                                             submission['is_iceberg_median'] # use the best\n",
    "                                            )\n",
    "                                   )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "submission[['id', 'is_iceberg']].to_csv('Submissions/submission_ensemble_resnet_cpl_oof_augment_minmax_median.csv'.format(repeat_count)\n",
    "                  , float_format=\"%.15f\", index=False) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "submission.to_csv('Data/submission_ensemble_resnet_cpl_oof_augment_origin.csv'.format(repeat_count)\n",
    "                  , float_format=\"%.15f\", index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "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
}