nn_regression_initialization.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Import necessary modules"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "from tensorflow.keras import layers, Model\n",
    "import numpy as np\n",
    "import matplotlib as mpl\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import StandardScaler"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load and prepare energy efficiency data set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!wget https://archive.ics.uci.edu/ml/machine-learning-databases/00242/ENB2012_data.xlsx\n",
    "\n",
    "data_df = pd.read_excel('ENB2012_data.xlsx')\n",
    "data = np.array(data_df)\n",
    "\n",
    "N = data.shape[0]\n",
    "X = data[:,:8]\n",
    "y = data[:,8]\n",
    "\n",
    "X_train_full, X_test, y_train_full, y_test = train_test_split(X, y, random_state=42)\n",
    "X_train, X_validate, y_train, y_validate = train_test_split(X_train_full, y_train_full, random_state=42)\n",
    "\n",
    "scaler = StandardScaler()\n",
    "X_train = scaler.fit_transform(X_train)\n",
    "X_validate = scaler.transform(X_validate)\n",
    "X_test = scaler.transform(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Construct FFNN model with sigmoid activation and normally distributed random weight initialization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         [(None, 8)]               0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 30)                270       \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 30)                930       \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 30)                930       \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 30)                930       \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 1)                 31        \n",
      "=================================================================\n",
      "Total params: 3,091\n",
      "Trainable params: 3,091\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(42)\n",
    "tf.random.set_seed(42)\n",
    "\n",
    "initializer_name = \"random_normal\"\n",
    "\n",
    "input_layer1 = layers.Input(shape=(8,))\n",
    "dense_layer1 = layers.Dense(30, activation=\"sigmoid\", kernel_initializer=initializer_name, input_shape=(8,))(input_layer1)\n",
    "dense_layer2 = layers.Dense(30, activation=\"sigmoid\", kernel_initializer=initializer_name)(dense_layer1)\n",
    "dense_layer3 = layers.Dense(30, activation=\"sigmoid\", kernel_initializer=initializer_name)(dense_layer2)\n",
    "dense_layer4 = layers.Dense(30, activation=\"sigmoid\", kernel_initializer=initializer_name)(dense_layer3)\n",
    "output_layer1 = layers.Dense(1, kernel_initializer=initializer_name)(dense_layer4)\n",
    "\n",
    "model = Model(inputs=input_layer1, outputs=output_layer1)\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Configure and run training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=\"mean_squared_error\", optimizer=keras.optimizers.SGD(lr=1e-3))\n",
    "history = model.fit(X_train, y_train, batch_size=1, epochs=20, validation_data=(X_validate, y_validate))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plot training history"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 1800x1200 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(dpi=300)\n",
    "pd.DataFrame(history.history).plot(figsize=(10, 5),logy=True)\n",
    "plt.grid(True)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Construct same FFNN model with Glorot initialization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         [(None, 8)]               0         \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 30)                270       \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 30)                930       \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 30)                930       \n",
      "_________________________________________________________________\n",
      "dense_8 (Dense)              (None, 30)                930       \n",
      "_________________________________________________________________\n",
      "dense_9 (Dense)              (None, 1)                 31        \n",
      "=================================================================\n",
      "Total params: 3,091\n",
      "Trainable params: 3,091\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(42)\n",
    "tf.random.set_seed(42)\n",
    "\n",
    "initializer_name = \"glorot_normal\"\n",
    "\n",
    "input_layer1 = layers.Input(shape=(8,))\n",
    "dense_layer1 = layers.Dense(30, activation=\"sigmoid\", kernel_initializer=initializer_name, input_shape=(8,))(input_layer1)\n",
    "dense_layer2 = layers.Dense(30, activation=\"sigmoid\", kernel_initializer=initializer_name)(dense_layer1)\n",
    "dense_layer3 = layers.Dense(30, activation=\"sigmoid\", kernel_initializer=initializer_name)(dense_layer2)\n",
    "dense_layer4 = layers.Dense(30, activation=\"sigmoid\", kernel_initializer=initializer_name)(dense_layer3)\n",
    "output_layer1 = layers.Dense(1, kernel_initializer=initializer_name)(dense_layer4)\n",
    "\n",
    "model = Model(inputs=input_layer1, outputs=output_layer1)\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Configure and run training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(loss=\"mean_squared_error\", optimizer=keras.optimizers.SGD(lr=1e-3))\n",
    "history = model.fit(X_train, y_train, batch_size=1, epochs=20, validation_data=(X_validate, y_validate))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plot (improved) training history"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 1800x1200 with 0 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlYAAAEvCAYAAACHYI+LAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nOzdd3zV1eH/8de592ZPIITkhiVDNiQRtGBFXDgQQUUFLaKtWmu169vWrp9d2tba1raOKrXWheAsCrhbcaEoIHuJKDNAwsgk697P74+TBSSQcW/uvcn7+fA+7s3n87nnnnwM5M2ZxnEcRERERKTtXKGugIiIiEhHoWAlIiIiEiAKViIiIiIBomAlIiIiEiAKViIiIiIBomAlIiIiEiCeUFcAIC0tzenbt29QP6O0tJSEhISgfkak0L2op3th6T7U072op3tRT/fC0n2wli9fXuA4TvfGzoVFsOrbty/Lli0L6mcsXryYCRMmBPUzIoXuRT3dC0v3oZ7uRT3di3q6F5bug2WM2dbUOXUFioiIiASIgpWIiIhIgChYiYiIiARIWIyxEhERkfZTVVXFzp07KS8vb9H7UlJS2LBhQ5BqFX5iY2Pp2bMnUVFRzX6PgpWIiEgns3PnTpKSkujbty/GmGa/r7i4mKSkpCDWLHw4jsP+/fvZuXMnJ510UrPfp65AERGRTqa8vJxu3bq1KFR1NsYYunXr1uJWPQUrERGRTkih6sRac48UrERERKTdJSYmhroKQaFgJSIiIhIgnSJYfVlQypLd1aGuhoiIiBzFcRx+9KMfMXz4cEaMGMEzzzwDQF5eHuPHjyc7O5vhw4fz3nvv4fP5uO666+quvffee0Nc+2N1ilmBD7/7OXNXV5DY4zNuO3uA+pVFRETCxIsvvsjKlStZtWoVBQUFjBkzhvHjx/P0009z/vnn8/Of/xyfz0dZWRkrV65k165drF27FoBDhw6FuPbH6hTB6jdThrN9Zx5/eXMzeYWH+e2U4XjcnaKxTkRE5Lh+vWAd63cXNetan8+H2+0+4XVDvcn8cvKwZpX5/vvvM2PGDNxuNz169ODMM8/kk08+YcyYMXz961+nqqqKqVOnkp2dTb9+/di6dSu33XYbkyZNYuLEic36jPbUKdJFlNvFDSOiufWsAcz9eAc3Pbmcskp1DYqIiISa4ziNHh8/fjzvvvsuWVlZzJw5kyeeeIIuXbqwatUqJkyYwAMPPMANN9zQzrU9sU7RYgV2yuQPzx9EZmos/2/+WqbP/ohHrxtDWmJMqKsmIiISMs1tWYLgLBA6fvx4Hn74YWbNmsWBAwd49913ueeee9i2bRtZWVnceOONlJaWsmLFCi666CKio6O5/PLL6d+/P9ddd11A6xIInSZY1brmtD70SIrl1rkruOzBJTz+9VM5KS0h1NUSERHplC699FI+/PBDRo0ahTGGP/7xj2RkZPD4449zzz33EBUVRWJiIk888QS7du3i+uuvx+/3A/D73/8+xLU/VqcLVgDnDu3BvJvG8vXHPuGyBz/gX9eNIbd3l1BXS0REpNMoKSkBbI/SPffcwz333HPE+VmzZjFr1qxj3rdixYp2qV9rdYoxVo3J7pXKi98aR3JcFFf/8yPeXL831FUSERGRCNdpgxVA37QEXvjWOAZlJPPNJ5fx5EfbQl0lERERiWCdOlgBpCXGMPfG0zhrUDr/b/5a7n5tY5MzFERERESOp9MHK4D4aA8PzzyFq0/rzT8Wf84Pnl1FZbU/1NUSERGRCNMpB683xuN2cdfU4XhTYvnTG5vZV1zOP752CsmxUaGumoiIiEQItVg1YIzh1rMH8qcrRrF06wGufOhD9hSWh7paIiIiEiEUrBox7ZSePHrdGHYcKOOyBz9g897iUFdJREREIoCCVRPGn9ydZ28eS5XfYdo/lvDR1v2hrpKIiEinlJiY2OS5L7/8kuHDh7djbY5Pweo4hnlT+M8t40hPjuXaf33MglW7Q10lERERCWMKVifQs0s8z988luxeqdw291MeeW9rqKskIiIS0W6//XYefPDBuq9/9atf8etf/5pzzjmH3NxcRowYwUsvvdTicsvLy7n++usZMWIEOTk5vP322wCsW7eOU089lezsbEaOHMlnn31GaWkpkyZNYtSoUQwfPpxnnnkmIN+bZgU2Q2p8NE9841R+8OxK7ly0gV2HDvOLSUNxu0yoqyYiItI2r/4E9qxp1qVxvmpwNyM6ZIyAC//Q5Onp06fzve99j1tuuQWAZ599ltdee43vf//7JCcnU1BQwFe+8hUuueQSjGn+79oHHngAgDVr1rBx40YmTpzI5s2beeihh/jud7/LNddcQ2VlJT6fj1deeQWv18uiRYsAKCwsbPbnHE/AW6yMMVONMf80xrxkjJkY6PJDJTbKzf0zcrn+9L78+4MvufXpFZRX+UJdLRERkYiTk5PDvn372L17N6tWraJLly5kZmbys5/9jJEjR3Luueeya9cu9u5t2XZz77//PjNnzgRg8ODB9OnTh82bNzN27Fh+97vfcffdd7Nt2zbi4uIYMWIEb731FrfffjvvvfceKSkpAfnemtViZYx5FLgY2Oc4zvAGxy8A/ga4gUccx/mD4zjzgfnGmC7An4A3AlLTMOByGX45eRhZqXHcuWgDBSVL+ee1o0mNjw511URERFrnOC1LRztcXExSUlJAPnbatGk8//zz7Nmzh+nTpzNnzhzy8/NZvnw5UVFR9O3bl/Lyli151NTOKVdffTWnnXYaixYt4vzzz+eRRx7h7LPPZvny5bzyyiv89Kc/ZeLEidxxxx1t/r6a22L1GHBBwwPGGDfwAHAhMBSYYYwZ2uCSX9ScD73CXSQVbQ5YcTec0Y/7r85h1Y5CLv/HEnYcKAtY2SIiIp3B9OnTmTdvHs8//zzTpk2jsLCQ9PR0oqKiePvtt9m2reX7944fP545c+YAsHnzZrZv386gQYPYunUr/fr14zvf+Q6XXHIJq1evZvfu3cTHx/O1r32NH/7wh6xYsSIg31ezgpXjOO8CB446fCqwxXGcrY7jVALzgCnGuht41XGcwNSyrd6+i9wVt8Nbv4bqioAUefFIL09841Tyiyu47B9LWLsrMH2zIiIincGwYcMoLi4mKyuLzMxMrrnmGpYtW8bo0aOZM2cOgwcPbnGZt9xyCz6fjxEjRnDVVVfx2GOPERMTwzPPPMPw4cPJzs5m48aNXHvttaxZs6ZuQPtdd93FL37xi4B8X6a5Gw4bY/oCC2u7Ao0x04ALHMe5oebrmcBpwGZgFvAJsNJxnIeaKO8m4CaAHj16nDJv3rw2fSPH464upc/Gh+ld8A6l8b3ZOPg7FCcPDEjZu4r9/Hl5OWVVDrfmxDA8LfznA5SUlBx3TZDORPfC0n2op3tRT/eiXke7FykpKQwYMKDF7/P5fLjd7iDUKHxt2bLlmIHtZ5111nLHcUY3dn1bUkBjw/Qdx3H+Dvz9RG92HGc2MBtg9OjRzoQJE9pQlRNb7Emgt/dWEhZ8h1M+vR3O+AGM/zF42j4+6pzx5Vz374/564oS/nD5YKad0jMANQ6exYsXE+z7HSl0Lyzdh3q6F/V0L+p1tHuxYcOGVo2VKg7gGKtIERsbS05OTrOvb8uswJ1ArwZf9wTCewXNkyfCLR/CyKvg3Xtg9gTYvbLNxWakxPLczWM5rV9XfvjcKuZ+vL3tdRUREZE6a9asITs7+4jHaaedFupqHaMtweoTYKAx5iRjTDQwHXg5MNUKorgucOk/YMYzULYfHjkH3v4dVFe2qdik2Cj+fd2p5PZO5eF3Pm9yZoKIiIi03IgRI1i5cuURj6VLl4a6WsdoVrAyxswFPgQGGWN2GmO+4ThONXAr8DqwAXjWcZx1watqgA26wLZeDb8c3rkb/nl2sxdIa0q0x8VVY3rx5f4y1u4qClBFRUREAk8NACfWmnvU3FmBMxzHyXQcJ8pxnJ6O4/yr5vgrjuOc7DhOf8dx7mrxp4dafFe4bDZMfxpK9tquwcV3g6+q1UWePyyDKLdhwerw7hUVEZHOKzY2lv379ytcHYfjOOzfv5/Y2NgWvS/8p7C1h8GToPdYePXHsPh3sHEhXPoQ9BjW4qJS46MZP7A7C1ft5icXDMalbW9ERCTM9OzZk507d5Kfn9+i95WXl7c4aESy2NhYevZs2YQ0Bata8V3h8kdg6BRY+H14+EyYcDuc/v3m7YvUwMWjMvnvxn2s2H6Q0X27BqnCIiIirRMVFcVJJ53U4vctXry4RTPkOqOA7xUY8YZMhluW2uf/3WkHt+9d36Iizh3SgxiPi4Wr84JUSREREQlHIQ1WxpjJxpjZgdpROmASusEV/4YrHofCHTD7THjvz+Crbtbbk2KjOHtwOgtX5+Hzq/9aRESkswhpsHIcZ4HjODcFakfpgBs21bZeDboQ/vsb+Nd5kL+pWW+dPMpLQUkFS7fuD3IlRUREJFyoK/BEErvDlU/AtH/DwS/hoTPgg7+B33fct501KJ2EaLdmB4qIiHQiClbNNfwy+PZSGHgevHkHPHo+FHzW5OVx0W7OHdqDV9fuocrnb8eKioiISKgoWLVEYjpc9RRc/i/YvwUe+iosua/J1qvJI70cKqvi/S0F7VxRERERCQUFq5YyBkZMs2Ov+p8Db/wC/n0hFGw55tIzTk4jOdbDglXqDhQREekMFKxaK6kHTJ8Dl862A9ofOh0+fBD89d1+MR43FwzP4I11eymvOv6YLBEREYl8ClZtYQyMugpu+Qj6TYDXfwqPXQSHttddcvFILyUV1Sze1LLVbUVERCTyKFgFQnImzJgHUx+CvNXwxv+rOzWufze6JkRrdqCIiEgnoGAVKMZA9gwYdAHsXFZ32ON2cdGIDP63YR9llc1bYFREREQik1ZeDzRvDhTthJJ9dYcmj/RyuMrHWxv2HeeNIiIiEum08nqgeXPt8+5P6w6N6duVHskxmh0oIiLSwakrMNAyRwLmiGDlchkmjfDyzqZ8Cg9Xha5uIiIiElQKVoEWkwTdB8GuFUccnjwqk0qfnzfW7QlRxURERCTYFKyCwZtjW6wcp+5Qdq9UenaJY+HqvBBWTERERIJJwSoYvLlQug+KdtUdMsYweZSX97cUcKC0MoSVExERkWBRsAoGb459bjDOCuzsQJ/f4dW1arUSERHpiBSsgiFjOLg8x4yzGpKZRL/uCZodKCIi0kEpWAVDVBykD4HdRwYrYwyTR3pZ+sUB9hWVh6hyIiIiEiwKVsHizT1mADvY2YGOA4vWqDtQRESko1GwCpasXCgvhANbjzg8ID2JIZnJ6g4UERHpgLSlTbA0MYAd4OKRmazYfogdB8rauVIiIiISTNrSJljSh4I7ptFgNXmkF1B3oIiISEejrsBgcUdBxohGg1XvbvGM6pXKwtXqDhQREelIFKyCKSsXdq8Ev++YU5NHZrJ2VxFb80tCUDEREREJBgWrYPLmQFUpFHx2zKmLR3oxBm1xIyIi0oEoWAWTN9c+H7WeFUBGSixj+nTl5VW7cY5akkFEREQik4JVMKUNhKiEY1ZgrzV5VCZb9pWwaW9xO1dMREREgkHBKphcbvBmNzqAHeDCEZm4DCxcpe5AERGRjkDBKti8ObBnDVRXHnMqLTGG0weksWC1ugNFREQ6AgWrYPPmgK8C8jc0evrikZls21/Gml0dcJFUERGRTkbBKtiyagawNzHO6vxhGUS5jba4ERER6QAUrIKty0kQm9rkOKvU+GjGD+zOwtV5+P3qDhQREYlkClbBZoztDmxkyYVak0d5ySssZ8X2g+1YMREREQk0bcLcHrw5sG8DVB1u9PS5Q3sQ43GpO1BERCTCaRPm9pCVC/5q2LO20dOJMR7OHpzOojV5VPv87Vw5ERERCRR1BbYHb459bmKcFdjuwIKSSpZ+caCdKiUiIiKBpmDVHpKzICH9uOOszhqUTkK0m4Wr1R0oIiISqRSs2oMxtjuwiSUXAOKi3Zw3tAevrt1DZbW6A0VERCKRglV78eZAwWaoaHpfwMmjvBwqq+KDLQXtWDEREREJFAWr9uLNBRzIW9XkJWcM7E5yrEezA0VERCKUglV7acYA9miPiwuGZ/DG+r2UV/naqWIiIiISKApW7SWxO6T0Ou44K7DdgSUV1SzelN9OFRMREZFAUbBqT97s47ZYAYzt141uCdEs0OxAERGRiKNg1Z68uXDwCyhreq0qj9vFhSMy+O+GvZRWVLdj5URERKStFKzaU+04q7yVx71s8kgv5VV+3tqwtx0qJSIiIoGiYNWeaoPVCcZZjenblR7JMSxYldcOlRIREZFAUbBqT3Gp0LX/CcdZuVyGi0d6eXdzPoWHq9qpciIiItJWClbtzZtzwmAFdnZgpc/PG+v2tEOlREREJBBCGqyMMZONMbMLCwtDWY32lZULRbug+Pjjp0b1TKFX1zgWrFZ3oIiISKQIabByHGeB4zg3paSkhLIa7asZC4UCGGO7Az/YUsD+kop2qJiIiIi0lboC21vGSDAu2H38AexgZwf6/A6vrlV3oIiISCRQsGpvMYmQNqhZ46yGZCbRv3sCC7VYqIiISERQsAqFrFy75ILjHPcyYwyTR3lZ+sUB9haVt1PlREREpLUUrELBmwNlBVC484SXXjzSi+PAIg1iFxERCXsKVqHgzbXPzRhnNSA9kSGZydo7UEREJAIoWIVCxnBwRTVrnBXA5FGZfLr9EDsOlAW5YiIiItIWClah4ImBHkNPuLVNrckjvQAsWqPuQBERkXCmYBUq3lzYvfKEA9gBenWNZ1SvVBasUnegiIhIOFOwChVvDlQUwoGtzbp88shM1u0u4vP8kiBXTERERFpLwSpUsmoGsDezO/DikV6MgYWr1B0oIiISrhSsQqX7YPDENnsAe0ZKLGP6dmXB6t04zeg+FBERkfanYBUq7ii7vU0zllyoNXmUly37Sti0tziIFRMREZHWUrAKJW8O5K0Cv69Zl184PAOXQYPYRUREwpSCVShl5UJVGeRvatblaYkxnD4gjQWr8tQdKCIiEoYUrELJm2OfmznOCuyaVtsPlLF6Z2GQKiUiIiKtpWAVSt0GQnRSi8ZZnT8sgyi3YaG2uBEREQk7Clah5HKBN7tFLVYp8VGMH9idhavz8PvVHSgiIhJOQhqsjDGTjTGzCws7cbeWNxv2rIHqyma/ZfIoL3mF5SzffjCIFRMREZGWCmmwchxngeM4N6WkpISyGqHlzQVfJexb1+y3nDu0BzEel2YHioiIhBl1BYZaKwawJ8Z4OGdIOq+syaPa5w9SxURERKSlFKxCrUtfiOvS7K1tak0e6aWgpJKPth4ITr1ERESkxRSsQs0Y22q1e2WL3nbW4HQSYzzMX7krSBUTERGRllKwCgfeXNi3HirLmv2W2Cg3Fw7P4LW1eyivat7K7SIiIhJcClbhwJsDjg/2rm3R26bmZFFSUc1bG/YGqWIiIiLSEgpW4SAr1z63cJzVV/p1o0dyDPM/VXegiIhIOFCwCgfJXkjMaNHMQAC3yzAlO4vFm/I5UNr8dbBEREQkOBSswoU3p0Vb29Samp1Ftd9h0Zq8IFRKREREWkLBKlxk5ULBZ1Be1KK3DclM4uQeieoOFBERCQMKVuHCmwM4kLeqRW8zxjA1J4vl2w6yfX/zZxWKiIhI4ClYhYu6Fdhb3h04JTsLQGtaiYiIhJiCVbhISIOU3i0ewA6QlRrHaSd1Zf7KXTiOE4TKiYiISHMoWIWTrJwWL7lQa2pOFlvzS1mzqzDAlRIREZHmUrAKJ94cOLQNylq+/99FwzOJdrv4jwaxi4iIhIyCVTjx1iwU2opxVinxUZw9OJ0Fq3ZT7fMHuGIiIiLSHApW4cSbbZ9bMc4KYGqOl4KSSj74fH8AKyUiIiLNpWAVTmJToNsA2NW6YDVhUDrJsR6taSUiIhIiClbhxpvb6har2Cg3k0Zm8vq6PZRVVge4YiIiInIiClbhxpsDxbuheE+r3j41O4uySh9vrt8b4IqJiIjIiShYhZus2gHsrWu1GtO3K96UWM0OFBERCQEFq3CTMQKMq9XrWblchik5Wbz3WQH5xRUBrpyIiIgcj4JVuIlOgO5DWrXkQq1Lc7Lw+R0Wrt4dwIqJiIjIiYQ0WBljJhtjZhcWarXwI3hzbFdgK7enOblHEkMzk5m/UsFKRESkPYU0WDmOs8BxnJtSUlJCWY3wk5UDZfvh0PZWFzE1x8uqHYfYml8SwIqJiIjI8agrMBx52zaAHeCSUVkYg1qtRERE2pGCVTjqMQxcUW0aZ5WREsu4/t2Y/+kunFZ2KYqIiEjLKFiFI08MZAxvU4sV2DWtth8o49MdhwJUMRERETkeBatw5c2B3SvB3/oNlS8YnkGMx6UtbkRERNqJglW48uZCRREc2NrqIpJiozh3aA8Wrs6jytf6gCYiIiLNo2AVrrw59rkN46wALs3O4kBpJe99lh+ASomIiMjxKFiFq+6DwRPX5nFW40/uTmp8FP/5VLMDRUREgk3BKly5PZA5stVb29SK9ri4eGQmb6zbQ3F5VYAqJyIiIo1RsApn3lzIWwW+6jYVc2lOFhXVfl5ftzdAFRMREZHGKFiFs6xcqD4MBZvaVExu7y706hrHSys1O1BERCSYFKzCWe0A9jZ2BxpjmJqdxQdbCthbVB6AiomIiEhjFKzCWdf+EJPc5gHsAFOys/A7sGCVBrGLiIgEi4JVOHO5IHNUm5dcABiQnsjInin8R4uFioiIBI2CVbjLyoU9a6G6os1FTc3OYt3uInaVaLFQERGRYFCwCnfeHPBXwd51bS7q4lGZuAx8uLttswxFRESkcQpW4c6ba58DMM4qPSmWrw7szoe7q/H7nTaXJyIiIkdSsAp3qb0hrmtAxlkBXJrjZX+5w/LtBwNSnoiIiNRTsAp3xthxVrva3mIFMHFoBtFuNIhdREQkCBSsIoE3B/I3QGVZm4tKiPGQm+5m0eo8Kqp9AaiciIiI1FKwigTeXHD8sGd1QIob5/VQeLiKxZvyA1KeiIiIWApWkaB2BfYADGAHGNbNTbeEaG1xIyIiEmAKVpEgOROSMtu8tU0tt8sweZSXtzbso/BwVUDKFBEREQWryOHNDViLFcDUnCwqq/28tjYvYGWKiIh0dgpWkcKbA/s/g/LCgBQ3qmcKJ6UlaHagiIhIAClYRYqsmnFWeasCUpwxhqnZWSz94gC7Dx0OSJkiIiKdnYJVpMisCVYBGmcFMCXbi+PAy6t2B6xMERGRzkzBKlIkdIPUPgEdZ9U3LYGc3qnMV3egiIhIQChYRRJvTsC2tql1aU4WG/cUsyGvKKDlioiIdEYKVpEkKxcObYfSgoAVOWlEJh6XYb7WtBIREWkzBatI4s21z7tXBqzIbokxjD+5Oy99uhu/3wlYuSIiIp2RglUkyRwFmIB3B07NyWJPUTkffbE/oOWKiIh0NgpWkSQ2GdIGBnQAO8B5Q3qQEO3mpU81O1BERKQtQhqsjDGTjTGzCwsDs+hlp+DNCeiSCwBx0W7OH57BK2vyKK/yBbRsERGRziSkwcpxnAWO49yUkpISympEFm8ulOyBosBuRXNpThbFFdX8b+O+gJYrIiLSmagrMNJ4axYKDfA4q3H90+ieFKMtbkRERNpAwSrSZIwA4w74OCu3yzBllJfFm/ZxqKwyoGWLiIh0FgpWkSY6HtKHBHycFdjZgVU+h0VrAtvNKCIi0lkoWEUib45tsXICu+7UMG8yA9ITtcWNiIhIKylYRaKsXDh8AA5tC2ixxhguzcniky8PsuNAWUDLFhER6QwUrCJR7QD2IHQHXjLKC8DLq7SmlYiISEspWEWi9GHgjg74AHaAXl3jGdO3Cy+u2IkT4K5GERGRjk7BKhJ5oqHH8KAEK7CD2D/PL2Xd7qKglC8iItJRKVhFqqxcuxmz3x/woieNyCTKbTSIXUREpIUUrCKVNwcqi2H/loAXnRofzYRB6by0ajc+v7oDRUREmkvBKlL1Os0+f/DXgC+7AHaLm/ziCpZ8XhDwskVERDoqBatIlTYQzrwdVs6B9/4U8OLPHpxOUoxHW9yIiIi0gIJVJJvwUxh5FfzvTljzfECLjo1yc9GITF5fu4fDlb6Ali0iItJRKVhFMmPgkvugz+kw/1uwbUlAi5+S46W00sebG/YGtFwREZGOSsEq0nli4KqnILUPzLsa9n8esKK/clI3MlNiNTtQRESkmRSsOoL4rnDNs2BcMGcalO4PSLEul+GSbC/vbs5nf0lFQMoUERHpyBSsOoqu/WDGPCjcZVuuqsoDUuylOVlU+x0WrckLSHkiIiIdmYJVR9LrVLjsYdjxEbx0S0AWDx2ckczgjCReXKHuQBERkRNRsOpohl0K5/4K1r4Ab98ZkCKnndKTlTsO8cLynQEpTyRsHfiCbgVLQ10LEYlgClYd0enfg9xZ8N6fYcWTbS7uunF9GduvGz/9zxpW7zwUgAqKhKn/3MyItb+D1c+FuiYiEqEUrDoiY2DSn6H/2bDwe/D5220qzuN2cf/VOXRPjOGbTy4nv1gD2aUD2vEJ7PiIKk8ivHyr3YtTRKSFFKw6KncUXPE4pA2CZ6+FfRvaVFy3xBgennkKB8sq+fbTK6jyBX7zZ5GQ+vA+iE1h+Sl/gfg0mHcNlOSHulYiEmEUrDqy2GS4+hmIioM5V0Bx2xb6HJ6Vwt2Xj+TjLw5w58L1AaqkSBg48AVsWACjv055XA+YPgfK9tt/lFRXhrp2IhJBFKw6utReNlyV7Ye5V0FlaZuKm5KdxQ1fPYnHP9zGc8t2BKiSIiH20T/AuOHUb9qvvdkw5X7YvgRe+0lo6yYiEUXBqjPw5sC0RyFvFbxwIzht2/vvJxcO5vQB3fj5/LWs2qHB7BLhyg7Ap0/BiGmQnFl/fMQ0GPcdWPYvWP5YyKonIpFFwaqzGHQhXPAH2LSI/p8/1qaiPG4X983I1WB26RiW/xuqSmHsrceeO/dXMOBcWPRD2P5Re9dMRCKQglVncto34bRv0Wvny/DxP9tUVNeEaGZfewqHDldyy5zlVFZrMLtEoOpKWDob+p0FGcOPPe9yw+WP2C71Z2banQ1ERCs6Ou8AACAASURBVI5DwaqzOf8uCrqdCq/+GDa91qaihnntYPZPvjzIbzWYXSLR2uehZA+Ma6S1qlZcF5g+F6rK4JmvBWy7KBHpmBSsOhuXm/VD/w8yRsLzX2/zWj1TsrO4aXw/nvxoG89+osHsEkEcB5bcD+lDof85x782fTBcNht2r7BrwzlO+9RRRCKOglUn5HfH2pmC8V3h6augsG1b1fz4/EF8dUAav5i/lk+3HwxQLUWC7PP/wb51dmyVMSe+fvAkmPAzWDXXziIUEWmEglVnlZQBVz9ruzeevgrKi1pdlB3MnkOPlBhufmo5+4rVVSIRYMl9kJhhZ/811/gfweCL4Y2ft3lHAxHpmBSsOrMeQ+HKxyF/Izx/PfiqW11Ul4RoHv7aaAoPV3HLUys0mF3C2561sPVtOO0m8MQ0/30uF1z6kN3R4Pnr7cKiIiINKFh1dv3Phkl/gS1vwSs/bNPYkaHeZP44bRTLth3kNwvXBbCSIgH24QMQFQ+nXN/y98YkwYyn7Z+VeVdDRUng6yciEUvBSuCUWfDV79v1fJb8vU1FXTLKyzfH9+Opj7Yz7+PtAaqgSAAV5cGa5yBnph1n2Bpd+8EVj9nW3vnf0mB2EamjYCXW2XfAsMvgzTtg3fw2FfXjCwZzxsA07nhpHSs0mF3CzccP290HvvKttpXT/yw477ew4WV490+BqZuIRDwFK7FcLpj6D+h1Gvznm7Djk1YX5XYZ7puRQ0ZKLDc/uZx9RRrMLmGiogSWPQpDJkPXk9pe3thvw8ir4O07YdOrbS9PRCKegpXUi4q1CyEmZcLc6W0amJsab1dmLy6v5ltzNJhdwsSnT0F5IYy9LTDlGQOT/2b343zhRsjfFJhyRSRiKVjJkRK6wTXP266Sp6+Ew63vyhuckcw9V4xk+baD/GqBBrNLiPmq4aMHbatsrzGBKzcqDq56yv7DZO4MOKyNyUU6MwUrOVbaAJj+NBz80u6PVl3Z6qIuHunl5jP78/TS7czVYHYJpY0L4NA2GBeg1qqGUnrClU/Coe3wwg3g9wX+M0QkIihYSeP6jIMpD8CX78HLt7Vp1tOPzh/E+JO7c8dLa1m+7UAAKynSTLXb13Q5CQZdFJzP6DMWLroHtrwJ//1NcD5DRMKegpU0beSVcNbPYfU8eOfuVhfjdhn+Pj0bb2ocNz+1gr0azC7tbcdS2LXMDjZ3uYP3OaOvh9Ffhw/+CmueD97niEjYUrCS4xv/I8i+Bhb/Ht7+fatbrlLjo5k9czSlFdXc/NRyKqrVVSLtaMl9ENfF/iwH2wV3Q++x8NKtkLcq+J8nImFFwUqOzxiY/HfI+Rq88wdY9INWjx8ZlJHEn64YxafbD/GrlzWYPSh2r9Qedkfb/zlsXASjvwHR8cH/PE80XPmEXXx03jVQWhD8zxSRsKFgJSfm9sAl99vV2Zc9Cs9dB1Wt6867aEQmt0zoz9yPdzBn6bbA1rOz2/Ex/PtCeOpy2Lo41LUJHx8+AO4oOPWm9vvMxHSYPgdK8+HZWeCrar/PFpGQUrCS5jEGzv0VnP97u9L0nGlQXtSqov5v4iAmDOrOr15ex7IvNZg9IPaugzlXQFIGpA204ffgl6GuVeiVHYCVT9vxgkk92vezvTlwyX2w7X14/Wft+9kiEjIKVtIyY2+By/4J2z+ExyZByb4WF+F2Gf52VQ7e1Di+NWcFewo1mL1NDnwBT15q11OaOd8uleH4bTdUZWmoaxdan/wLqg/D2FtD8/kjr7Sf/fFsWPFkaOogIu1KwUpabuSVMOMZ2L8F/jWxVSu0p8RHaTB7IBTvgSengq8SZv4HuvSBbv1h2r9h3/rOvUFwVbndF3DAeZA+JHT1OPfX0P9sOz6xDVtFiUhkULCS1hl4LsxaAOWHbLjKW93iIgZlJPGXK0excsch7pi/DqezBoDWKjtgW6pK8uGaF44MDwPOsb/Q178E73XSDYLXPGvHOI0LUWtVLbcHpj0KyVnwzNegKC+09RGRoFKwktbrORq+/jq4o2234BfvtbiIC4ZncutZA3hm2Q6eWqqV2ZutstRuObR/C8x4Gnqecuw1426DEVfC/+6CTa+1fx1Dye+3C4JmjICTzgx1bexSDzPmQmUJPHNNqyd/iEj4U7CStuk+CL7xut24+anLYf3LLS7i++edzFmDuvPrl9fxiQazn1h1hR0/tWu5bQnpN6Hx64yBS/4OmaPsNiv5m9uzlqG15S0o2GQ3WzYm1LWx0ofApQ/b/2+LftB5u2hFOjgFK2m7lJ7w9dcgcyQ8NwuWP9ait7tdhr9Oz6FX13i+9dQKtu8vU7dgU/w+ePEm2Pq2nXE2ZPLxr4+Ks9P+o2JhXifaIPjD+yDJC8MvC3VNjjTkYjjzJ7ByDix9ONS1EZEg8IS6AtJBxHeFa1+ya/Ys+K4d9zP+h81uLUiJi2L2zFOY+sAHjL/nbdwuQ3Ksh9T4aJLjokiJiyK19jnePic3OJYSH0VqXDQpcVHERrkw4dJKEUiOAwu/D+vnw8S77KKtzZHS0y5Y+fhkePFGmDEvuNu6hFreKvjiXTjvN3b9qnBz5u2wd61dgqE4D2JTwBNju9Q9MeCOsYuMHvEc08i5hu+JDp+WOZFOTsFKAic6wY4jeelWePtOKN1nt/dwNa9hdGCPJJ67eRwfbCmg8HAVhYerOFTzXFhWyfb9pXXH/cdp0Ip2u0iJPzKM1YavlKPC2ZeHfPQ/UEa3xGjio8P8j8Nbv4IVj8MZP2z5gOw+4+DCP9ouqP/dCef+MihVDAtL7ofoRMidFeqaNM7lgksfshMPPvhr4Mp1Hx3GGjxHJdjxZn3G2UdSRuA+V0SOEPDfJMaYfsDPgRTHcaYFunwJc+4omPoPSEiDD++Hsv0w9SH7l3szDPUmM9SbfNxr/H6HkspqCsuq6gNYw9eHKylqcGxPUTkb9xRTdLiK4orqY8r7zUd2C5i4KDfdEqPplhhDt4Ro+6h9nXjk664J0cR42rHV5/177S/h0d+As3/RujLGfAP2rIb3/wIZw2H45YGtYzgo3AlrX4DTboa41FDXpmkxSXDDW3aQva/CjpvzVR71XAHVlUc9H31defPeW1FkF0r95J/287v2g97joM9Yu69h135q8RIJkGYFK2PMo8DFwD7HcYY3OH4B8DfADTziOM4fHMfZCnzDGKOt3TsrlwvOv8tu6/HmHXZZgKuetL9MAlK8ITk2iuTYKHq18L3VPj9F5dUUHq7iYFkl7360HG+/QewvqWR/SQUHSispKK1kb1E5G/KK2F9SSaXP32hZSbGeo8JX4yGsW0IMXeKj8LhbOaRx+WO2tWr4NLjoT237BXjhPbBvI8z/NnQbaMfFdSRLH7LPX7k5tPVoLpcLXHF2LFyw+apssN62BLZ9CJtegZVP2XOJGTUhq6ZFK31os1uaReRIzW2xegy4H3ii9oAxxg08AJwH7AQ+Mca87DjO+kBXUiLU6d+F+DR4+TY7vuea521LVgh53C66JtgWp5NIoGirhwmjm45njuNQXFHN/pJKDpRWUFBSeeTrUvt6+4EyVmw/xMGySnyN9FN6XIYB6YkMzbQtckMzkxmSmUyXhBO05K19ERZ8DwZOtN1Hbf1lV7tB8OwJdmbhTW+H/P9JwJQXwfLHYdhUSO0d6tqEH3cUZJ1iH+Nus61lBZth2wd2J4VtS2Ddf+y1sSnQ6yv08mXAjnjIzG52q7NIZ9esYOU4zrvGmL5HHT4V2FLTQoUxZh4wBVCwkno519iB7c9dB4+eD1970a4OHiGMqW8dOykt4YTX+/0OhYer2F9aYVvBSm1L2O7CcjbmFfHB5wW8+OmuuuszU2KPCFtDvcn06hKPy2XskgEv3gS9vwJXPB64gdhJPexMwX9faP+/zPxPYMoNtRVP2C6vUG1fE2lcLkgfbB9jvmEnRxzaXh+yti2h//7XYevj4Imz69b1GWe7DnudasdUisgxTHOntdcEq4W1XYHGmGnABY7j3FDz9UzgNOCXwF3YlqxHHMf5fRPl3QTcBNCjR49T5s2b16Zv5ERKSkpITEwM6mdEilDci+TCDYxY81v8rhhWj/wlpYl92/XzmxKKe1FU4bC92M+OYj/bi3xsL/aTV+rUDciPdcPEhC3cU3kXB6Mz+e+g39I9NYlod2DHwPTYs5ghG+9lZ9YkVmZeHdF/Poy/mtOW3kx5bDorc37XprL0d0W9yoO78FZvJ/XQOlIK15NY8gUGPw4uipP6U5gyjEOpQylMGUJ11PHHRkY6/VxYug/WWWedtdxxnNGNnWvL4PXG/pZ3HMfZD5xwgIPjOLOB2QCjR492JkyY0IaqnNjixYsJ9mdEitDciwnwlTPhqcsYs/YOu9dgn7HtXIdjhcvPRXmVj817i+24rs9XcO2mP7LH6cLlRbdT8Ikblymjf/fEI1q2hmQmk5YY04ZPnQCvV9Hzw/spSezH4IvvDNS30/7WPA8V+cRedh8TBk1oU1Hh8jMRDhYvXszwCdfUHygvgp0fY7YtIXnbhyTvepVeO+fbc92H2D/TAyfCyRd0uMHwHe7nonCnXRevhT0IHe4+BEFbgtVOOGLscE9gd9uqIx1aj6HwjTfsNPMnp9qNggdfFOpahYXYKDcje6YyMv4AvPNjSEwm/vpXecHpzvrdRWzIK2J9XhGffHGAl1bW/zFLT4o5ImwNzUzGmxqH33Hw+W0rmN/v4HdqXjtO3TnHAV/Oj0nfuZqBm//B9lXnUpqea6/x22t9joPjOPhqvvbXlOmrKSfG7SI22k18tJv4KA+x0S7ioz3ERblxu9rpF6vjwJL77GD8gee3z2d2VrHJMOBcfP3OoaSimtLSEnw7luPe+RGxu5eSvPIZPMsepXLUTKIn/0XjssLR3vV2hvGa5yEqHma+aLt2JWDaEqw+AQYaY04CdgHTgasDUivpuFJ72/0F51xhN6S95O/NX+iyoyvKgyem2tlb17+Mq2tf+gB9uiVw4YjMussOllbWBa31eUWs313E+58VUH28xb2OI4Wv8XL0RmJfnMW0irvYR5eAfDvRHhfx0W7iotzE1YQv+9pDfM2xuJpj8ce89tS9jq15Toj2EB9TX07dIrDbPoC8lXDxXzWTrQmO41BR7bdhqKK65tlX9/rI49WU1JzbtrucBzd9SOlRxw9X+Y76hBwgBzc+vu95nltXPcnG9cv5cPS9nDpiCEMzkzvmor2RZPtSu2zL5lftuman3gSfvQFPXgZfewF6nxbqGnYYzV1uYS4wAUgzxuwEfuk4zr+MMbcCr2OXW3jUcZx1QaupdBwJaTBrgQ1WL30bSvPh9O91uK6DFik7YFvyyvbDrJftgOImdEmIZtyANMYNqJ/NV1Ht47O9JazPK6KgpAK3MbiMweUyuIzdNsgYU3OcmuP1595c+VOu2/Fz3ug2m08mPAmeWNwuGrzH2JUBjMFdU6YxhspqP4erfByu9FFW6at5Xd3gdc25Bq8LD1exp/DwMe9ryS5GxkB8lJv4GA9/8/+eoSaFWz7ui2f1xzXHGwSxKA8JMW7ioz22ZS3aTUJM7WvPEV/HeI4NZk5NC5+vpiXPV9saWHuswXG/v+G1Tt21jsORx2teV/scKqv9VPr89rnaT0WD1/acr8FrPxVHnGv8dUUj5xqbrdqYuCh7PxJj3DhVDj0SoUdyLIkxnrrj9tlT91z7OiHGTXH56by8ZAwTt/yWlCUzuGnxD9iTOITxA7szYVB3zhiYRmq8WrLahePYSTDv32v/ARLXBSb81Iaq+K5w+nfgsYvhqcvsxCKFq4Bo7qzAGU0cfwV4JaA1ks4hJhGufhbm32zXaCrJh4l3ds4Wh4oSePpKOPC5XZIi65QWFxHjcTM8K4XhWSmtqsLiQ/3wnPowqc/O5Lytf4Qp97dr0K1tUWkYzg5X+imrrK4LZWWVPvt1pY+yimpKK30kFH3OuM3LeDn1WoiKo6gmtJVW1F9bUd34OmSNcRnwGODNV+tCUai5XYZot4toT83D7SLGc+TX0R4XibEeot0uojwuYtzHnj8yDLlJjIkiIcbdIBR5SIh2H7Hemh1P04qxkP1ug7wzSZ97NfNLfstT3f6Pezfm8MKKnbgMZPdK5cyT0zlzUHdGZqXYWbASOL5qu/XV+3+FvWsgOQsu+APkXnvkbM5kL1y3CB6bpHAVQGG+h4d0aJ5ouOwRu9bVRw9AWQFMeSA893cLluoK23K3azlc+ST0OzN0dRl6CYz/Mbz7R7tw6GnfbLePNsYQG2W7/Vrk5YfAE8slN9zBJYndG72k2uevC2elFTZslVbYwFZW4aO0stqeq6ymrMLHli+20bdPb9wubGtdTeueu+65vuXOXdfy1/jxI59tS6G7wfEotzkmAEV7XMS43XWv222sWqBljsT9zcXw7CxmbfsdM8fexspB32PxZwd4Z3M+f/3vZu59azNdE6I5Y2BaTWtW9zZOyOjkqsph1dPwwd/h4BeQdjJMeRBGXNH0eLfkTBuuHq9tuXrBLvEiraZgJaHlcsGFd0Nid7uHXdl+u2ZTTCeYzuv3wQs3wNa37V9+Qy4OdY1sN8HetfDaTyF9CJw0PtQ1alpJPqyaB9kz7M9PEzxuF8luF8mxzQvsixfnMWFC012x0gIJaXDtfHj9Z7g+vI/cfevInfYoPzjvZA6UVvLeZ/m8symfdzbn103KGJGVwoRB3Tnz5O5k90pt/Y4FLVBe5SOvsJxdBw+z61AZuw7Z17sPHWbXocPsKSrH5fhJXfJfkuM8JMVGkRRrn5Nj679OjvWQHFd/ruE1CdGe4LXMlRfBskfhowehZC94c2Hib2HQpOb1AiRnwqyFNeHqcttyHgaztiOVgpWEnjEw/keQ0B0Wfh9+nwUuD3hi7caynljwxNQ8H/V13fnaYzENro2p2YQ25qgyao/F4akqDs337Diw4Luw4WU4//d2IdVw4HLBpQ/DI+fCs7PgpsXhu6DrJ4/YvfC0IGh4c0fBRfdAj+Gw6P9g9lkwYy5d04cwJTuLKdlZ+P0O63YX8c7mfSzelM8Db2/hvv9tITnWwxkDbcgaf3J3MlJiW/zxjmMX7d3ZICjVPtsgVU5BScUR7zEGeiTF4k2NZVSvVC5IieXLbdtJSUujqLyK4nK7G8OXBaUUl1dTVF5Fle/43cbGQGKMh+S6EFYbvI4MY8mxdpP41PgousRH0yU+mtT4qMZbc0vyYek/4ONHoKIQ+p0Fl/3T/oOopV35tS1Xj02COdMUrtogpMHKGDMZmDxgwIBQVkPCxSnXQbcB8OUHdnPZuk1my+1GsrUbztaeqyiu//ro8/6qZn3kVwHWnwTeHMjKtc+ZowK2r2GT3rwDPn3Sdr2NvSW4n9VSsckwYy788yy77c03Xg+/VbarDtsNhU++ENIGhro20hynzILug23X9yPnwmWzYfAkwHaRjuiZwoieKdx69kAKy6p4f0sB72zexzub81m0Jg+AwRlJnDmoOxNOTueUPl2I9rio9vnZV1zRICg1CE81Yaq08shZjDEeF1mpcWR1iWNwRjJZXeLwpsaRlRpHzy5x9EiOJfqoiQyLF+9lwoRRjX5rtWMEa0NX0WH7bB9VdeGr4XNxeRV5heVs3ld/7fHG9MVGuegSH01KXBQDow9wecWLnF70Kh6nii+6n8PWQTfgeHPp4ooiNb/UhrO4Fu5RmpRRE65qWq6+9oLCVSuENFg5jrMAWDB69OgbQ1kPCSN9v2ofbeX329aMpkJZdTlUlvL5RwvpH3MIdi6DdS/WvNnYsQnenPrAlTEicBvlvn8vLPk7jLkRzvpZYMoMtG794fJH4ekr4KVbYdqj4TVrc9Vc2208Tq1VEaX3abYV9JlrYN7VMOFntrX6qO6qlPgoJo3MZNLITBzHYeOeYt7ZbLsNH33/Cx5+ZysJ0W5S46PZU1R+TCDpEh9FVpc4TkpL4KsD02yIqglS3tQ4uiVEB3T5h4ZjBNNb+W8yx3E4XGVnzR4qs5vEF5ZVcbDm9aGySmIPbOT0PU8x5tDb+HGxyJzJ/ZWT2LIjA3b4gWXHlJsU66lr9UqNj6ZLTeBKjY8mPtqNg12jrnZWruM4xA24j0tXf5PExy9l/rC/sSs5B8ee5IttlSyr2ISDfY9jD+Ngv3BqyvA7DY4D0W5X3T2KjXIRV/f6yK/jot3EetzERtvr46LcRLVDd3AgqStQOiaXC1xxJwxDO3bH0L92FeHSAtj9qX3sWmHHPq2u2WrJuCF9KHiz68NW+rCWL4C47N92FuSIK+DCP4ZXWDnawHPhnF/CW7+0wfKMH4S6RpbfDx8+YP8/9Dk91LWRlkrJgutftZuLL/6dnbU29aEmx1UaYxhSs2n5zWf2p6SimiVbCnhncz6lFdVkdYkjKzUeb2osPWuCU3x05P1qM8bULP/hITPlqL+3tn9k/0H2+Wt2Daqxt+Ae+22mJnuZUrNRfGFNADtYVsWhssq6cHao5uva49v2l3KwtJKi8urj1uchfszc6LuYtPo2rqu8nU8cO+7QAGbrFowx9rUBg6HmP1zG1Byj7hqgbqmQ1nC7TE3wOl44cxNXc/7GM/rRq2t8qz4rECLvp08kWBLSYOB59lGraPeRYWvjItuFB3Z8V4/hR7ZspQ0CdxN/rNa+YMeQDTwfpv4jMpaWOP27sGcN/Pc30GMYnBwGK5tvfg32b4HL/xXewVSaFhUHlz5kZ5++8Qv413kw/WnoetIJ35oY42HisAwmDstoh4qGkOPAZ2/aQLV9CcR1hbN+DmNusGtQ1Wi4UXxLwkR1TdCpDUK2LBuS6oJR6bm4npjMs4V/xlzzHPQ9vU1b2vj9tsu0vMouq1L/bI81/PpwlY+Kmtm85dX1x8rrHn4OV/ooqagmv7iCimp/3bVXju51xLYw7U3BSuR4kr32UTMWBMeBQ9th94r6sLXmOVj2L3veE2d/WXhz68NW1/7w+f/gxW9C77FwxWORs6SEMXDJfVCw2c5gvPF/oR/T9OH9kNILhk4NbT2kbYyBsd+2s0+fu96O6bviMeg3IcQVC7G6NajutTN0k3vCBXdD7syAjnX0uF0nHn+VnFE/W3DOFXDNc236TJfL1O24EJj9HcKTgpVISxhjZ8l16QPDLrXH/H67uGfDlq3lj9nZOgDRSXZcV/pguHoeRIeuibpVouNh+hyYPQHmzoAb/wuxrVuItM12LbcrSJ//u6ZbBiWy9D8bbnob5l5tt1eZeCd85VsdvzXSVw1Fu+DQNvuPtdrHtiX2WNog27I94orQ/kMsqUdNuJoMc6aRMuwX2I1YpCn6m0mkrVwu24qTNhBGXmmP+aqhYFN92Koosb8wQhVI2iq1N1z5BDwxBV68CabPDU1X5pL7ISYZcma2/2dL8HTtBze8Cf+5GV7/qe1+vvheiGr58gphw++zQwkObT8yPB2seV20C5wGsxWNC5K89u+R838Hgy4Kn+ECST3guoXw2MWMXP1ryB4VmElGHZSClUgwuD12TFKPYR1nk+m+X7XbYrzyQ3j7Ljjn/7Xv5x/cButfst1Hscnt+9kSfDFJdveBd+6Gd/5g/2Fy1Ry7vlI48vugeM9RwWlbfXgq2gX+hgPEDSRl2n+k9Blrn1N7Q2of+5yc1fLJMO0pMR2uW0j5g2eRMOcKuyXZSWeEulZhScFKRJpvzA2wZzW89yc7U3BYO45zWvqQ7R467eb2+0xpXy4XnPVTyBhuxyTOPhOuegp6nRqa+jiO3Rpm96f03vZfePmFBt12O45dLy8xw4aknmOgy7Qjw1NKT7swcSRLTGdl9l2c/tnv7f6mCleNUrASkeYzBi76E+zbCPO/Zf+FnpgBCd3sno8J3e3sykD/Ajl8CFY8AcMvt9P1pWMbMhlu6A/zZtiVwC++t31afkvy7cSUXcvrH4cPAtAPYF+6DUqZ2TDkEjvWsi449YrsrstmqopOhVkL7JgrhatGaeV1EWkZTwxc9ST8+yK7gnxjopNs2EroXhO4jgpe8Wn2ufb1iX4hLX8MKku0fU1n0mMo3Pg2PH89vPRtyFsN598VuIHclaWQt+rIEHVouz1nXHbduiGTIesU8Oby7vo8xp8TBsuNhIPE9KNmCz4b3vuKtjOtvC4iLZeUAbcth/JCuwJ6aQGU5kNZgX1dtt9+XVoARTshb6V93dRWQ9GJRwauo4PX0oftX9yZI9v3+5TQiu8K17xgA/xHD8C+9XaT9oRuLSvHVw35G2sC1DI7c3ffenBqFqxM6W2XRhlzow1SmaOOWbDUv+lAgL6pDiKxe4PZglcqXDWgrkARaR1jIC7VPrr1P/H1jgMVRTUhrKAmhOU3EsR22daJ0vwjg9iU+4L3vUj4cnvggt/ZMX0Lvgv/nGBnpWYMb/z62rXm6lqiVthgX1Vmz8em2vA06CL7nJVrW2Ck5RK7227BJy6x4erqZ6DfmaGuVcgpWIlI+zDGLjcRm9LyIOarsuuASeeVPcPu4fnMNXal9qkP2rXkyg7Y8NSwS6+swL7HHWNbOXNn1Yeorv06/hpZ7SmxO1z7sg1XT19l1+rrNyHUtQopBSsRCU8Ng5gIQM9TajZxngnPXQdv3AGFNeOiMNB9kN12KSvXBqnW7OcpLVfbcvX45Jpw9UynDlcKViIiEjmSMuxilf+70y6FMObrNeOisrW+WSglpNWEq0s6fbhSsBIRkcjiiYGJvw11LeRoCWkw6+X6cDVjHvQ/K9S1andhsl6+iIiIRLzacNW1P8ydbjeg72QUrERERCRwarsFu/a3G7d3snClrkAREREJrIRu9UsxzJ1hV87v0vfIbX7iunTIGZoKViIiIhJ4Cd3sUgwv3girn7XLpzQUndQgaDV41G4VFJsakcFLW9qIiIhIcCR0g5kv2teHD9n9Res2sm7w+PJ9qCw+8r0xyY0Erz71r+NS2//7aQZtaSMiIiLBV7tTQ+aoY885jt3wx/XuuAAABkJJREFUurHQdXAbfPGu3S+0oZiUxlu6+oyz3Ywhoq5AERERCS1j7N6Q8V3Bm33s+brg1UiL14GtsHUxVJXaa2/4n11MNkQUrERERCS8HRG8co497zh2e6ND26B7aLe/UrASERGRyGaMHc+V0C3UNdE6ViIiIiKBomAlIiIiEiAKViIiIiIBomAlIiIiEiAKViIiIiIBomAlIiIiEiAKViIiIiIBomAlIiIiEiAhDVbGmMnGmNmFhYWhrIaIiIhIQIQ0WDmOs8BxnJtSUlJCWQ0RERGRgDCO44S6Dhhj8oFtQf6YNKAgyJ8RKXQv6uleWLoP9XQv6ule1NO9sHQfrD6O43Rv7ERYBKv2YIxZ5jjO6FDXIxzoXtTTvbB0H+rpXtTTvaine2HpPpyYBq+LiIiIBIiClYiIiEiAdKZgNTvUFQgjuhf1dC8s3Yd6uhf1dC/q6V5Yug8n0GnGWImIiIgEW2dqsRIREREJqg4XrIwxFxhjNhljthhjftLIeWOM+XvN+dXGmNxQ1DPYjDG9jDFvG2M2GGPWGWO+28g1E4wxhcaYlTWPO0JR12AzxnxpjFlT8z0ua+R8Z/mZGNTg//VKY0yRMeZ7R13TYX8mjDGPGmP2GWPWNjjW1RjzpjHms5rnLk2897h/r0SaJu7FPcaYjTV/Bv5jjElt4r3H/fMUaZq4F78yxuxq8Ofgoibe22F+Lpq4D880uAdfGmNWNvHeDvUz0WaO43SYB+AGPgf6AdHAKmDoUddcBLwKGOArwNJQ1ztI9yITyK15nQRsbuReTAAWhrqu7XAvvgTSjnO+U/xMHPU9u4E92LVYOsXPBDAeyAXWNjj2R+AnNa9/AtzdxL067t8rkfZo4l5MBDw1r+9u7F7UnDvun6dIezRxL34F/PAE7+tQPxeN3Yejzv8ZuKMz/Ey09dHRWqxOBbY4jrPVcZxKYB4w5ahrpgBPONZHQKoxJrO9KxpsjuPkOY6zouZ1MbAByAptrcJWp/iZOMo5wOeO4wR7Yd6w4TjOu8CBow5PAR6vef04MLWRtzbn75WI0ti9+P/t3M2rDXEcx/H3t7DwkEKelYesLDwkeV5ZICkWIuUWpVvuws7C/2AjUSKlmyRPd3HVLXsSeYywc3M7ysJNNi5fi9/v1GnMMJe553Tm93nVdGbm9731m1/fmb4zv5nr7kPuPhY3HwCL296xDijIizJqlRd/GgczM+AAcK2tnepSdSusFgEfWraH+b2YKBNTK2a2FFgLPMxp3mRmz8zsnpmtamvH2seBITN7bGbHc9qTywngIMUXyRRyommeu49AuBkB5ubEpJgfRwlPcfP87Xyqi744LXq5YIo4pbzYBjTc/V1Beyo5UUrdCivL2Zf97LFMTG2Y2XTgJnDS3UczzU8IU0GrgbPAnXb3r022uPs6YBdwwsy2Z9pTy4kpwF7gRk5zKjkxHqnlx2lgDOgvCPnb+VQH54EVwBpghDANlpVSXhziz0+rUsiJ0upWWA0DS1q2FwMf/yGmFsxsMqGo6nf3W9l2dx91969xfRCYbGZz2tzNCefuH+PvJ+A24RF+q2RyItoFPHH3RrYhlZxo0WhO+8bfTzkxyeSHmfUAe4DDHl+eySpxPnU9d2+4+w93/wlcJP8Yk8gLM5sE7AeuF8WkkBPjUbfC6hGw0syWxbvyg8BAJmYAOBK/BNsIfGlOBdRJnBO/BLx29zMFMfNjHGa2gZAPn9vXy4lnZtPMbEZznfCC7stMWBI50aLw7jOFnMgYAHrieg9wNyemzHWl65nZTuAUsNfdvxXElDmful7mHct95B9jEnkB7ADeuPtwXmMqOTEunX57vuqF8IXXW8LXGqfjvl6gN64bcC62vwDWd7rPEzQOWwmPpZ8DT+OyOzMWfcArwtcsD4DNne73BIzD8nh8z+KxJpsT8VinEgqlmS37ksgJQjE5AnwnPG04BswG7gPv4u+sGLsQGGz529+uK928FIzFe8I7Q83rxYXsWBSdT928FIzF1XgteE4olhbUPS/yxiHuv9K8PrTE1jon/nfRf14XERERqUjdpgJFREREOkaFlYiIiEhFVFiJiIiIVESFlYiIiEhFVFiJiIiIVESFlYiIiEhFVFiJiIiIVESFlYiIiEhFfgEcNSH9DmtM7wAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 720x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(dpi=300)\n",
    "pd.DataFrame(history.history).plot(figsize=(10, 5),logy=True)\n",
    "plt.grid(True)\n",
    "plt.show()"
   ]
  }
 ],
 "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}