project_lstm_keras.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import LSTM, Dense, Dropout, TimeDistributed, BatchNormalization\n",
    "import numpy as np\n",
    "import h5py\n",
    "import matplotlib.pyplot as plt\n",
    "from data_utils import get_data\n",
    "from sklearn import preprocessing\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train: (2508, 1000, 22) \n",
      "y_train: (2508,) \n",
      "X_val: (50, 1000, 22) \n",
      "y_val: (50,) \n",
      "X_test: (0, 1000, 22) \n",
      "y_test: (0,) \n"
     ]
    }
   ],
   "source": [
    "# Load data from all .mat files, combine them, eliminate EOG signals, shuffle and \n",
    "# seperate training data, validation data and testing data.\n",
    "# Also do mean subtraction on x.\n",
    "\n",
    "data = get_data('project_datasets',num_validation=50, num_test=0, subtract_mean=True, subtract_axis=1,\n",
    "                transpose=True)\n",
    "for k in data.keys():\n",
    "    print('{}: {} '.format(k, data[k].shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "# substract data from list\n",
    "X_train = data.get('X_train')\n",
    "y_train = data.get('y_train')\n",
    "X_val = data.get('X_val')\n",
    "y_val = data.get('y_val')\n",
    "#X_test = data.get('X_test')\n",
    "#y_test = data.get('y_test')\n",
    "\n",
    "# get data dimension\n",
    "N_train, T_train, C_train = data.get('X_train').shape\n",
    "N_val, T_val, C_val = data.get('X_val').shape\n",
    "#N_test, T_test, C_test = data.get('X_test').shape\n",
    "\n",
    "# add dummy zeros for y classification\n",
    "lb = preprocessing.LabelBinarizer()\n",
    "lb.fit([0, 1, 2, 3,])\n",
    "lb.classes_\n",
    "y_train = lb.transform(y_train)\n",
    "y_val = lb.transform(y_val)\n",
    "#y_test = lb.transform(y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "# # normalize data across the time sequence dimension only\n",
    "# scaler = preprocessing.MinMaxScaler()\n",
    "# for i in range(N_train):\n",
    "#     scaler.fit(X_train[i,:,:])\n",
    "#     X_train[i,:,:] = scaler.transform(X_train[i,:,:])\n",
    "    \n",
    "# for i in range(N_val):\n",
    "#     scaler.fit(X_val[i,:,:])\n",
    "#     X_val[i,:,:] = scaler.transform(X_val[i,:,:])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train:  (2508, 250, 22)\n",
      "X_val:  (50, 250, 22)\n"
     ]
    }
   ],
   "source": [
    "# construct X_total and y_total based on sub-sampling of X_train and y_train\n",
    "\n",
    "# take sub-sampling on the time sequence to reduce dimension for RNN\n",
    "sampling = 4\n",
    "\n",
    "X_train = X_train.reshape(N_train,int(T_train/sampling), sampling, C_train)[:,:,0,:]\n",
    "X_val = X_val.reshape(N_val,int(T_val/sampling), sampling, C_val)[:,:,0,:]\n",
    "    \n",
    "# get new data dimension\n",
    "N_train, T_train, C_train = X_train.shape\n",
    "N_val, T_val, C_val = X_val.shape\n",
    "\n",
    "print('X_train: ', X_train.shape)\n",
    "print('X_val: ', X_val.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Expected input batch shape: (batch_size, timesteps, data_dim)\n",
    "# Note that we have to provide the full batch_input_shape since the network is stateful.\n",
    "# the sample of index i in batch k is the follow-up for the sample i in batch k-1.\n",
    "\n",
    "# perhaps should try masking layer\n",
    "\n",
    "data_dim = C_train\n",
    "timesteps = T_train\n",
    "num_classes = 4\n",
    "batch_size = 50\n",
    "num_epoch = 30\n",
    "\n",
    "model = Sequential()\n",
    "model.add(LSTM(40, return_sequences=True, stateful=False,\n",
    "          batch_input_shape=(batch_size, timesteps, data_dim)))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(BatchNormalization())\n",
    "model.add(LSTM(40, return_sequences=True, stateful=False))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(BatchNormalization())\n",
    "model.add(LSTM(40, stateful=False))\n",
    "model.add(Dropout(0.5))\n",
    "# model.add(TimeDistributed(Dense(T_train)))\n",
    "model.add(Dense(50,activation='softmax'))\n",
    "model.add(Dense(num_classes, activation='softmax'))\n",
    "\n",
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adam',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 2500 samples, validate on 50 samples\n",
      "Epoch 1/30\n",
      "2500/2500 [==============================] - 20s 8ms/step - loss: 1.3863 - acc: 0.2576 - val_loss: 1.3869 - val_acc: 0.2000\n",
      "Epoch 2/30\n",
      "2500/2500 [==============================] - 16s 6ms/step - loss: 1.3856 - acc: 0.2528 - val_loss: 1.3884 - val_acc: 0.1800\n",
      "Epoch 3/30\n",
      "2500/2500 [==============================] - 16s 7ms/step - loss: 1.3845 - acc: 0.2848 - val_loss: 1.3875 - val_acc: 0.2200\n",
      "Epoch 4/30\n",
      "2500/2500 [==============================] - 16s 6ms/step - loss: 1.3815 - acc: 0.3092 - val_loss: 1.3876 - val_acc: 0.2600\n",
      "Epoch 5/30\n",
      "2500/2500 [==============================] - 16s 7ms/step - loss: 1.3782 - acc: 0.3064 - val_loss: 1.3881 - val_acc: 0.2600\n",
      "Epoch 6/30\n",
      "2500/2500 [==============================] - 16s 6ms/step - loss: 1.3727 - acc: 0.3212 - val_loss: 1.3871 - val_acc: 0.3200\n",
      "Epoch 7/30\n",
      "2500/2500 [==============================] - 19s 7ms/step - loss: 1.3672 - acc: 0.3360 - val_loss: 1.3878 - val_acc: 0.2400\n",
      "Epoch 8/30\n",
      "2500/2500 [==============================] - 19s 8ms/step - loss: 1.3607 - acc: 0.3392 - val_loss: 1.4135 - val_acc: 0.2800\n",
      "Epoch 9/30\n",
      "2500/2500 [==============================] - 19s 8ms/step - loss: 1.3601 - acc: 0.3332 - val_loss: 1.4029 - val_acc: 0.2600\n",
      "Epoch 10/30\n",
      "2500/2500 [==============================] - 19s 7ms/step - loss: 1.3545 - acc: 0.3436 - val_loss: 1.4079 - val_acc: 0.2600\n",
      "Epoch 11/30\n",
      "2500/2500 [==============================] - 17s 7ms/step - loss: 1.3484 - acc: 0.3512 - val_loss: 1.3931 - val_acc: 0.3800\n",
      "Epoch 12/30\n",
      "2500/2500 [==============================] - 17s 7ms/step - loss: 1.3420 - acc: 0.3656 - val_loss: 1.3860 - val_acc: 0.3000\n",
      "Epoch 13/30\n",
      "2500/2500 [==============================] - 16s 7ms/step - loss: 1.3310 - acc: 0.3708 - val_loss: 1.3712 - val_acc: 0.3400\n",
      "Epoch 14/30\n",
      "2500/2500 [==============================] - 16s 7ms/step - loss: 1.3278 - acc: 0.3816 - val_loss: 1.4131 - val_acc: 0.2600\n",
      "Epoch 15/30\n",
      "2500/2500 [==============================] - 16s 7ms/step - loss: 1.3226 - acc: 0.3756 - val_loss: 1.4146 - val_acc: 0.2600\n",
      "Epoch 16/30\n",
      "2500/2500 [==============================] - 19s 7ms/step - loss: 1.3232 - acc: 0.3796 - val_loss: 1.4168 - val_acc: 0.2200\n",
      "Epoch 17/30\n",
      "2500/2500 [==============================] - 19s 8ms/step - loss: 1.3151 - acc: 0.3892 - val_loss: 1.3943 - val_acc: 0.2800\n",
      "Epoch 18/30\n",
      "2500/2500 [==============================] - 19s 8ms/step - loss: 1.3107 - acc: 0.3952 - val_loss: 1.4206 - val_acc: 0.2200\n",
      "Epoch 19/30\n",
      "2500/2500 [==============================] - 19s 8ms/step - loss: 1.3081 - acc: 0.4076 - val_loss: 1.4247 - val_acc: 0.2600\n",
      "Epoch 20/30\n",
      "2500/2500 [==============================] - 19s 7ms/step - loss: 1.3154 - acc: 0.3912 - val_loss: 1.3850 - val_acc: 0.4000\n",
      "Epoch 21/30\n",
      "2500/2500 [==============================] - 16s 7ms/step - loss: 1.2928 - acc: 0.4056 - val_loss: 1.4195 - val_acc: 0.2800\n",
      "Epoch 22/30\n",
      "2500/2500 [==============================] - 17s 7ms/step - loss: 1.2809 - acc: 0.4252 - val_loss: 1.5148 - val_acc: 0.2000\n",
      "Epoch 23/30\n",
      "2500/2500 [==============================] - 17s 7ms/step - loss: 1.2821 - acc: 0.4232 - val_loss: 1.4257 - val_acc: 0.3000\n",
      "Epoch 24/30\n",
      "2500/2500 [==============================] - 17s 7ms/step - loss: 1.2784 - acc: 0.4420 - val_loss: 1.5354 - val_acc: 0.2200\n",
      "Epoch 25/30\n",
      "2500/2500 [==============================] - 16s 6ms/step - loss: 1.2767 - acc: 0.4196 - val_loss: 1.4633 - val_acc: 0.3000\n",
      "Epoch 26/30\n",
      "2500/2500 [==============================] - 16s 7ms/step - loss: 1.2707 - acc: 0.4308 - val_loss: 1.3918 - val_acc: 0.3600\n",
      "Epoch 27/30\n",
      "2500/2500 [==============================] - 16s 7ms/step - loss: 1.2656 - acc: 0.4372 - val_loss: 1.4298 - val_acc: 0.3200\n",
      "Epoch 28/30\n",
      "2500/2500 [==============================] - 17s 7ms/step - loss: 1.2596 - acc: 0.4416 - val_loss: 1.4263 - val_acc: 0.3600\n",
      "Epoch 29/30\n",
      "2500/2500 [==============================] - 16s 7ms/step - loss: 1.2592 - acc: 0.4448 - val_loss: 1.5231 - val_acc: 0.2000\n",
      "Epoch 30/30\n",
      "2500/2500 [==============================] - 17s 7ms/step - loss: 1.2468 - acc: 0.4492 - val_loss: 1.5091 - val_acc: 0.2800\n"
     ]
    }
   ],
   "source": [
    "# train the data with validation\n",
    "history = model.fit(X_train[0:int(N_train/batch_size)*batch_size, :, :],\n",
    "                    y_train[0:int(N_train/batch_size)*batch_size, :],\n",
    "                    batch_size=batch_size, epochs=num_epoch, shuffle=False,\n",
    "                    validation_data=(X_val, y_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['val_loss', 'val_acc', 'loss', 'acc'])\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYsAAAEWCAYAAACXGLsWAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvNQv5yAAAIABJREFUeJzs3Xd4VGX2wPHvSUghEFoIPZDQexIITbqiYqEpSLEX7Cu6dlddy+q6urqWtaGiqFTBRRCRnygoSJEAAUInEEhCCyUQSvr7++NOcAiBTJKZTMn5PM88ZO7ccibROXPf8xYxxqCUUkpdjJ+7A1BKKeX5NFkopZQqkSYLpZRSJdJkoZRSqkSaLJRSSpVIk4VSSqkSabJQChCRL0TkHw7umywig1wdk1KeRJOFUkqpEmmyUMqHiEgVd8egfJMmC+U1bM0/j4vIBhE5JSKfiUh9EVkgIpkiskhEatvtP1RENolIhogsEZF2dq/Fisha23EzgOAi17pWRBJsxy4Xkc4OxniNiKwTkRMikiIiLxR5vY/tfBm212+zba8qIm+KyB4ROS4iy2zbBohIajG/h0G2n18QkVki8rWInABuE5HuIrLCdo39IvJfEQm0O76DiPwkIkdF5KCIPCMiDUTktIiE2e3XRUTSRSTAkfeufJsmC+VtrgcuB1oDQ4AFwDNAONZ/zw8BiEhrYBrwsO21H4B5IhJo++CcA3wF1AG+sZ0X27GxwCTgHiAM+BiYKyJBDsR3CrgFqAVcA9wnIsNt521mi/c9W0wxQILtuH8DXYFLbDE9ARQ4+DsZBsyyXXMKkA88AtQFegGXAffbYggFFgE/Ao2AlsDPxpgDwBLgBrvz3gxMN8bkOhiH8mGaLJS3ec8Yc9AYkwYsBVYZY9YZY7KA/wGxtv1GA/ONMT/ZPuz+DVTF+jDuCQQAbxtjco0xs4DVdte4G/jYGLPKGJNvjJkMZNuOuyhjzBJjzEZjTIExZgNWwupve3kcsMgYM8123SPGmAQR8QPuACYYY9Js11xujMl28Heywhgzx3bNM8aYNcaYlcaYPGNMMlayK4zhWuCAMeZNY0yWMSbTGLPK9tpk4CYAEfEHxmIlVKU0WSivc9Du5zPFPK9u+7kRsKfwBWNMAZACNLa9lmbOnUVzj93PzYBHbc04GSKSAUTYjrsoEekhIottzTfHgXuxvuFjO0dSMYfVxWoGK+41R6QUiaG1iHwvIgdsTVOvOhADwHdAexGJwrp7O26M+aOMMSkfo8lC+ap9WB/6AIiIYH1QpgH7gca2bYWa2v2cArxijKll9wgxxkxz4LpTgblAhDGmJvARUHidFKBFMcccBrIu8NopIMTuffhjNWHZKzp19IfAVqCVMaYGVjOdfQzNiwvcdnc2E+vu4mb0rkLZ0WShfNVM4BoRucxWoH0UqylpObACyAMeEpEAEbkO6G537CfAvba7BBGRarbCdagD1w0FjhpjskSkO1bTU6EpwCARuUFEqohImIjE2O56JgFviUgjEfEXkV62Gsl2INh2/QDgWaCk2kkocAI4KSJtgfvsXvseaCgiD4tIkIiEikgPu9e/BG4DhqLJQtnRZKF8kjFmG9Y35PewvrkPAYYYY3KMMTnAdVgfikex6hvf2h0bD4wH/gscA3ba9nXE/cBLIpIJPI+VtArPuxe4GitxHcUqbkfbXn4M2IhVOzkK/AvwM8Yct53zU6y7olPAOb2jivEYVpLKxEp8M+xiyMRqYhoCHAB2AAPtXv8dq7C+1hhj3zSnKjnRxY+UUvZE5BdgqjHmU3fHojyHJgul1Fki0g34CavmkunueJTn0GYopRQAIjIZawzGw5ooVFF6Z6GUUqpEemehlFKqRD4z6VjdunVNZGSku8NQSimvsmbNmsPGmKJjd87jM8kiMjKS+Ph4d4ehlFJeRUQc6iKtzVBKKaVKpMlCKaVUiTRZKKWUKpHP1CyKk5ubS2pqKllZWe4OxeWCg4Np0qQJAQG6To1Syvl8OlmkpqYSGhpKZGQk504w6luMMRw5coTU1FSioqLcHY5Sygf5dDNUVlYWYWFhPp0oAESEsLCwSnEHpZRyD59OFoDPJ4pCleV9KqXcw6XJQkQGi8g2EdkpIk8V8/ptthXFEmyPu+xeu1VEdtget7oyTqWU8lZz1qXx7dpUXD11k8uShW1Fr/eBq4D2wFgRaV/MrjOMMTG2x6e2Y+sAfwd6YC1K83cRqe2qWF0pIyODDz74oNTHXX311WRkZLggIqWUr8jKzefVH7Ywa02qy1sXXHln0R3YaYzZZVtsZjowzMFjrwR+MsYcNcYcw5oyebCL4nSpCyWLvLy8ix73ww8/UKtWLVeFpZTyAbPWpHIoM5sHB7Z0+bVcmSwac+5C8qm2bUVdLyIbRGSWiESU5lgRuVtE4kUkPj093VlxO9VTTz1FUlISMTExdOvWjb59+zJ06FDat7dusoYPH07Xrl3p0KEDEydOPHtcZGQkhw8fJjk5mXbt2jF+/Hg6dOjAFVdcwZkzZ9z1dpRSHiI3v4APlyTRpWkterUIc/n13N11dh4wzRiTLSL3AJOBSx092BgzEZgIEBcXd9EGuxfnbWLzvhPlifU87RvV4O9DOlx0n9dee43ExEQSEhJYsmQJ11xzDYmJiWe7uE6aNIk6depw5swZunXrxvXXX09Y2Ll/+B07djBt2jQ++eQTbrjhBmbPns1NN93k1PeilPIu3yXsIy3jDC8P71AhHVxceWeRBkTYPW9i23aWMeaIMSbb9vRToKujx3qr7t27nzMW4t133yU6OpqePXuSkpLCjh07zjsmKiqKmJgYALp27UpycnJFhauU8kD5BYYPFu+kfcMaDGxTr0Ku6co7i9VAKxGJwvqgH4O1iPxZItLQGLPf9nQosMX280LgVbui9hXA0+UJpqQ7gIpSrVq1sz8vWbKERYsWsWLFCkJCQhgwYECxYyWCgoLO/uzv76/NUEpVcgsS97Pr8Ck+uLFLhXWbd1myMMbkiciDWB/8/sAkY8wmEXkJiDfGzAUeEpGhQB5wFLjNduxREXkZK+EAvGSMOeqqWF0pNDSUzMziV6g8fvw4tWvXJiQkhK1bt7Jy5coKjk4p5W2MMfz3l520CK/G4A4NKuy6Lq1ZGGN+AH4osu15u5+f5gJ3DMaYScAkV8ZXEcLCwujduzcdO3akatWq1K9f/+xrgwcP5qOPPqJdu3a0adOGnj17ujFSpZQ3+HnLIbYeyOTNUdH4+VXcYFyfWYM7Li7OFF38aMuWLbRr185NEVW8yvZ+lapsjDGM+GA5R05l88ujAwjwL3/ZWUTWGGPiStrP56f7UEopX/H7ziMkpGRwb/8WTkkUpaHJQimlvMR/F++gfo0gRnZtUuHX1mShlFJeID75KCt3HeXufi0IquJf4dfXZKGUUl7gv4t3UqdaIGO7R5S8swtoslBKKQ+XmHacJdvSubNPFCGB7pl4Q5OFUkp5uP/+spPQ4Crc3KuZ22LQZOFhqlev7u4QlFIeZPvBTH7cdIDbLomkRnCA2+LQZKGUUh7sg8U7CQn05/beUSXv7ELunnXW5z311FNERETwwAMPAPDCCy9QpUoVFi9ezLFjx8jNzeUf//gHw4Y5utSHUsqTFBQYlicdIS6yNsEBzu2ltOfIKeau38edfaKoUy3QqecurcqTLBY8BQc2OvecDTrBVa9ddJfRo0fz8MMPn00WM2fOZOHChTz00EPUqFGDw4cP07NnT4YOHarraCvlhf65YAufLN1Ns7AQ/jmiE5e0rOu0c3+4JIkq/n6M79vcaecsK22GcrHY2FgOHTrEvn37WL9+PbVr16ZBgwY888wzdO7cmUGDBpGWlsbBgwfdHapSqpS++H03nyzdzTWdGgIw7tNVPDFrPRmnc8p97n0ZZ5i9NpXRcRHUqxFc7vOVV+W5syjhDsCVRo0axaxZszhw4ACjR49mypQppKens2bNGgICAoiMjCx2anKllOf6v00HePH7zQxqV593x8aSm1/AOz/vYOJvu/hl6yFeGNqBazo1LHOLwcTfdmEM3NPf/XcVoHcWFWL06NFMnz6dWbNmMWrUKI4fP069evUICAhg8eLF7Nmzx90hKqVKISElg4emr6Nzk1q8NzYWfz8hOMCfJwe3Ze6DvWlUqyoPTl3HXZPj2Zfh+Pozxhi2H8zko1+TmPbHXkbENqZJ7RAXvhPHVZ47Czfq0KEDmZmZNG7cmIYNG3LjjTcyZMgQOnXqRFxcHG3btnV3iEopB+09cpo7v1hNeGgQn90aR9XAc4vaHRrV5Nv7LuGL5cm8+X/bufytX3licFtu6tkM/2KmFM/KzWd50mEWb03nl62HSLMll06NazJhUKsKeU+O0GRRQTZu/LO4XrduXVasWFHsfidPnqyokJRSpXTsVA63ff4H+cbwxe3dqVs9qNj9qvj7cVff5lzZoQF/m5PI3+du4ruENF67vjOt64eSeuw0i7ce4peth1iedITsvAKqBvjTu2VdHhjYkoFtw2lYs2oFv7uL02ShlFIOyMrNZ/yX8aRmnGHKXT1oEV7yANqIOiFMvr0bcxLSeGneZq55dylN64SQlH4KgKZ1QhjbvSkD29ajR1Qdp3e9dSZNFkopVYKCAsOj36wnfs8x/jsulm6RdRw+VkQYEduEfq3CeWPhNtIyzpxNEM3rVvOaLvM+nyyMMV7zxygPX1nxUClP9K8ftzJ/w36eubot13ZuVKZzhFUP4rXrOzs5sorj072hgoODOXLkiM9/kBpjOHLkCMHB7u+LrZSv+XJFMh//totbejXziMFx7uLTdxZNmjQhNTWV9PR0d4ficsHBwTRpUvGrZynlyxZtPsgLczcxqF09/j6kQ6VopbgQn04WAQEBREW5d/ItpZR3Wp+SwV+mraNj45q8axtLUZn5dDOUUkqVxeJth7jx01WEVQ/ks1u7uW3BIU+iyUIppex8uSKZO79YTbOwEGbdewnhocWPpahsNF0qpRSQX2B4+fvNfLE8mUHt6vPOmBiqBelHZCH9TSilvN6XK5LZlX6K8f2a07hW6Uc+n8zO46Fp6/hl6yHu6hPF01e3q/Q1iqJc2gwlIoNFZJuI7BSRpy6y3/UiYkQkzvY8UkTOiEiC7fGRK+NUSnmvD5ck8fx3m/hieTID3ljMc3MS2X/c8cn70jLOMPLD5fy6PZ1XRnTk2Wvba6IohsvuLETEH3gfuBxIBVaLyFxjzOYi+4UCE4BVRU6RZIyJcVV8Sinv9/GvSfzrx60Mi2nEY1e04cNfk5i+ei8zVqcwrkdT7hvQgvoXWQtiQ2oGd06OJysnn89v60a/1uEVGL13ceWdRXdgpzFmlzEmB5gOFLd26MvAvwBd0EEp5bBPl+7inwu2MiS6EW+OiiaiTgivjujEL48O4Loujfl65R76vb6YF+dt4lDm+R8vPybu54aPVxBUxY/Z91+iiaIErkwWjYEUu+eptm1niUgXIMIYM7+Y46NEZJ2I/CoifYu7gIjcLSLxIhJfGQbeKaUsny3bzT/mb+GaTg35zw3RVPH/86Msok4Ir13fmV8eHcCwmEZ8ucJKGv/4fjPpmdkYY/jo1yTu/Xot7RrWYM4DvWldP9SN78Y7uK3ALSJ+wFvAbcW8vB9oaow5IiJdgTki0sEYc8J+J2PMRGAiQFxcnG/P6aGUAqylTF/+fjNXdWzA22NizkkU9pqGhfD6yGjuH9CS937ZyaTfdzNl1V6iI2qyctdRru3ckH+PivbomV49iSvvLNKACLvnTWzbCoUCHYElIpIM9ATmikicMSbbGHMEwBizBkgCWrswVqWUF/hyRTIvzNvMlR2spUwDLpAo7EXWrcabN0Sz6K/9GdyxAfHJx3hwYEveHROriaIUxFWT7IlIFWA7cBlWklgNjDPGbLrA/kuAx4wx8SISDhw1xuSLSHNgKdDJGHP0QteLi4sz8fHxzn4bSikP8fXKPTw7J5HL29fn/XFdCKxStu+62Xn5BFXRJFFIRNYYY+JK2s9lzVDGmDwReRBYCPgDk4wxm0TkJSDeGDP3Iof3A14SkVygALj3YolCKeXbpq7ay7NzErmsbb1yJQpAE0UZuezOoqLpnYVSvmnG6r08OXsjA9uE89HNXfXD3skcvbPQuaGUUh5rZnwKT327kf6tw/nwJk0U7qTJQinlkX5MPMCTszfQp2VdPr65qxaj3UyThVLK46xPyeDhGeuIiajFJ7fEaaLwAJoslFIeJS3jDHd9GU/d6kFMvFkThafQWWeVUh4jMyuXO79YTVZOPlPv6qFrSXgQTRZKKY+Ql1/Ag1PXsePQSb64vRutdAoOj6LNUEoptzPG8OK8zfy6PZ1/DO9I31Y6qZ+n0WShlHK7z39P5quVe7inX3PGdm/q7nBUMTRZKKXcatHmg7w835rv6cnBbd0djroATRZKKbdJTDvOQ9PX0alxTd4eHYufrlDnsTRZKKXc4sDxLO6cvJpaVQP49JY4qgZqF1lPpr2hlFIV7lR2HndOXs2p7Hy+ubcX9S6y9KnyDHpnoZSqUPkFhgnT17Fl/wneGxdLu4Y13B2ScoDeWSilXMIYQ3pmNnuPnj7nse1AJpv2neDlYR0Y2Kaeu8NUDtJkoZQqt7z8AmatSWXbwUz2HrGSQsqx02TlFpzdRwQa1ggmok4Iz13bnpt7RbovYGcyBrbMgzZXgX+Au6NxGU0WSqlyycsv4OEZCXy/YT/VAv2JqBNCVN1q9G8dTtOwEJrWsR6Na1f1zSnGdy2GmTfDiIkQPdrd0biMJgulVJnlFxge/WY932/Yz1NXteWefs0RqWTdX3cvtf5N/k2ThVJKFZVfYHjsm/V8l7CPJwa34d7+LdwdknskLzv3Xx+lvaGUUqWWX2B4fNZ6/rcujceuaM39A1q6OyT3yD4J+9ZCtXA4lgwZKe6OyGU0WSilSqWgwPDk7A18uzaNRwa15sFLW7k7JPdJWQkFedB7gvXch+8uNFkoVYlk5ebz5YpkRn20nLcXbSfjdE6pji8oMDz97UZmrUllwmWtmDCoEicKsJKDXxXoehtUre3TyUJrFkpVAiez8/h65R4+XbqbwyezaRYWwtuLdvDJb7u4qWcz7uwbRb3Qi4+iLigw/G1OIjPiU/jLpS15uLInCrCSQ+OuEBQKzXpD8lJ3R+QymiyU8mHHTuXwxfJkvliezPEzufRpWZcHBsbSs3kdth7I5IMlSXyydBefL09mdFwE9/RvTpPaIeedxxjD83MTmfbHXu4f0IK/Xt668vV6Kir7JKSthT4PW88j+8LW7yFjL9TyvWnWNVko5YMOncji02W7+XrlHk7n5HN5+/o8MLAlMRG1zu7TrmEN3hsby18vb81HS5KYvnov0/7Yy/DYxtw3oAUtwqsDVqL4+9xNfL1yL/f0b87jV7bRRAGwdyWYfCtJAET2sf5N/h1iNFkopTxYytHTfPxbEjPjU8nLL2BIdCPuH9CSNg0uvERpVN1q/GtkZyYMasXE33YxffVeZq9N5eqODbl/YAu+iU/lyxV7uLtfc54a3FYTRaHkpeAXABHdref12kPVOlbTVMxY98bmAposlPIBxhj+89N23l+ShJ/AyK5NuLd/C5qFVXP4HI1qVeWFoR148NKWTFq2m69W7GH+xv0A3Nkniqev0kRxjsJ6RaDtd+znB5G9rcF5PsilyUJEBgPvAP7Ap8aY1y6w3/XALKCbMSbetu1p4E4gH3jIGLPQlbEq5c3e+mk77/2ykxGxjXlicBsa1qxa5nPVrR7EE4Pbck//FkxZtQc/kco5MvtisjNh3zro88i52yP7WvNEHdsDtZu5JzYXcVmyEBF/4H3gciAVWC0ic40xm4vsFwpMAFbZbWsPjAE6AI2ARSLS2hiT76p4lfJW7y/eyXu/7GRMtwheHdHJaavN1awaUHkH25Vk7ypbvaLPudsLn+/53eeShSvHWXQHdhpjdhljcoDpwLBi9nsZ+BeQZbdtGDDdGJNtjNkN7LSdTyll57Nlu3lj4TaGxzTiFScmilIrKID1MyD3jHuuX9HO1it6nLs9vN2fdQsf48pk0RiwH/ueatt2loh0ASKMMfNLe6zt+LtFJF5E4tPT050TtVJe4uuVe3j5+81c3akB/x4Vjb8716/eNh/+dzesn+a+GCpS8lJoEgeBRboZn61b+N54C7eN4BYRP+At4NGynsMYM9EYE2eMiQsPD3decEp5uG/iU3h2TiKXta3H26NjqeLv5skYEqZa//rgN+rzZJ2AfQnnN0EViuxrjbU4tqdi43IxV/4XlgZE2D1vYttWKBToCCwRkWSgJzBXROIcOFapSmvu+n08OXsDfVvV5f0buxBYxc2J4uQh2L4QxM9KFsa4Nx5XS7lAvaJQ4bgLH0ucrvyvbDXQSkSiRCQQq2A9t/BFY8xxY0xdY0ykMSYSWAkMtfWGmguMEZEgEYkCWgF/uDBWpbzCwk0HeGRGAnGRdZh4cxzBAR6wmNDGb6wPzx73wsmDcGSnuyNyrcJ6RZMLlFHD20JImCYLRxlj8oAHgYXAFmCmMWaTiLwkIkNLOHYTMBPYDPwIPKA9oVRlt3jbIR6cupbOTWoy6bZuVA30gERhDKybYo036HaXtc0H2+vPkbys+HpFIT8/2zxRmiwcZoz5wRjT2hjTwhjzim3b88aYucXsO6BwjIXt+Su249oYYxa4Mk6lPN3ynYe596s1tGkQyhe3d6d6kIeMp92/Hg5tgphxUKc5hDb8c+U4X1RSvaJQZF847lt1C52iXCkPF598lDsnxxMZVo2v7uhBzaoB7g7pTwlTwT8IOl4PItaHpC/XLYrOB3UhZ+eJ8p27C4eShYh8KyLX2HowKaUqyM5Dmdz++Woa1gzm67t6ULtaoLtD+lNeNmycCW2vsdZyAOtD8tQhOLzDvbG5SvJS8A+EJt0uvp8P1i0c/fD/ABgH7BCR10SkjQtjUkoBZ3LyuX/KWgKr+PH1XT0IDw1yd0jn2r4QzhyDmBv/3Hb2G7WPNkUlL4PGF6lXFPLBuoVDycIYs8gYcyPQBUjGmn5juYjcLiIedE+slO94/rtEdhw6ydtjYmhUq+xzPblMwhSrRtFi4J/b6jSH0EY+9SF5VtYJ2O9AvaJQVD9b3SLZpWFVFIeblUQkDLgNuAtYhzVBYBfgJ5dEplQlNmtNKt+sSeUvA1vSt5UHDjjNPAg7foLOo8HPrleWiPVh6ot1i70rwRQ4nix8rG7haM3if8BSIAQYYowZaoyZYYz5C1DdlQEqVdnsOJjJc3MS6dm8DhMGtXZ3OMXbONMq9No3QRU6W7fYXvFxuVJhvSLCwWnqfKxu4Wj/u3eNMYuLe8EYE+fEeJSq1E7n5HH/lLVUC/Ln3TGx7p3v6UKMsXpBNekG4cUks6jCEcxLIdyHypvJS633HOBgk2DRuywvn+Ld0Wao9iJydj1GEaktIve7KCalKq3n5mxiZ/pJ3h4dS70awe4Op3j71sGhzdbYiuLUjoIajX3mGzUAWcetMSWONkEViuwLx1Mgw/vHWziaLMYbYzIKnxhjjgHjXROSUpXTN/EpzF6byl8ubUWfVnXdHc6FFY6t6HBd8a/7Yt2itPWKQj5Ut3A0WfiL3TJZtoWNPKjDt1Lul5NXwMe/JpGYdrzUx24/mMlz3yXSq3kYEy5r5YLonCQv25oLqt21ULXWhfeL7AOn0n2nbuHo+IqiwttCSF2fSBaO1ix+BGaIyMe25/fYtimlbP6zaDsfLkkCYEh0Ix69vDWRdUteA/tUtlWnqB4UwDtjYzyzTlFo2wLIyii+sG3PfryFL9QtkpeVrl5RqPAua/dSr69bOHpn8SSwGLjP9vgZeMJVQSnlbVYnH+WjX5O4rktjHhzYkkWbDzLorV95ds5GDp3IuuBxxhiem5NIUvpJ3h0TQ71QD61TFEqYao2jaD7g4vsV1i18YZ6ostYrCkX2gROpXj/ewqE7C2NMAfCh7aGUspOZlcsjMxJoWieEl4d1pFpQFW65pBnv/byTaX/sZdaaVO7oHcU9/VucN6/TN/GpfLsujYcHteKSlh5cpwDIPAA7f4LeD587tqI4hfNE7Vzk9d+o2bPCVq8oYT6oC7Ff36JOlPPiqmCOjrNoJSKzRGSziOwqfLg6OKW8wUvzNrMv4wxv3RBDNdtssPVCg3l5eEd+frQ/V3ZowAdLkuj3+mI+/jWJrFxrtv2tB07w3HeJ9G4Zxl8u9eA6RaENM6wPzQv1gioqsg+cPgzp21wbl6slL7UK+qWtVxQKb+MTdQtHm6E+x7qryAMGAl8CX7sqKKW8xY+JB/hmTSoPDGxJ12a1z3u9WVg13hkTy/yH+hDbtBb/XLCVAW8s4euVe3hgylpqVA3g7dEeOp7C3tmxFd2hroOJzVfmiTpbryhjE6GP9A5zNFlUNcb8DIgxZo8x5gXgGteFpZRrbd53gv5vLObz33eX+RyHMrN45n8b6dS4Jg+V0IOpQ6OafHF7d6bf3ZNGtYJ5dk4iuw+f4p0xMZ43QWBx9q2F9K0QW0Jh217tSKjRxLu/UZ/JgAMbyl6vKOQDdQtHe0Nl26Yn3yEiD2Kth63TfCivdORkNuO/jOfAiSxenLeZQ5nZPHFlG6QU7erGGJ6ctYFT2Xn8Z3Q0Af6Ofe/q2TyM2fddwuJth8jLN1zSwsPrFIXWTYEqwdBhhOPHFH6j9ua6RVnHVxQVaTeq3UvrFo7eWUzAmhfqIaArcBNwq6uCUspVcvMLuH/KWtJPZjPr3l7c2KMpHy5J4vFZG8jNL3D4PFP/2Mvibek8fVVbWtYLLVUMIsKlbetzRYcGpQ3fPXKzIHEWtBsCwTVLd+zZusVW18TmauWtVxQKbwPVwr36LqvEOwvbALzRxpjHgJPA7S6PSikXeXHeJlbtPsrbo2OIbVqbmIha1AsN5j+LtnP0VA7vj+tS4trWuw+f4h/fb6Fvq7rc0iuyYgJ3p20/WN1HHS1s24uy6wlUr51z46oIyUutiQPLWq8o5APzRJV4Z2GMyQfKeQ+mlPt9vXIPX6/cyz39mzM8tjFgfcufMKgVr4zoyJJthxj36UqOncq54Dny8gt4ZEYCgVX8eGNkNH6eXpj4EMVbAAAgAElEQVR2hoSpVu0hqn/pj63VDGpGeGeR+0wG7HdCvaJQZB84kQbHyl4ncydHm6HWichcEblZRK4rfLg0MqWcaOWuI7wwdxMD24TzxJVtz3v9xh7N+ODGrmzad4KRHy0nLeNMsed5f3ESCSkZvDKiIw1qevgAOmc4sR+SfoboMSWPrSiON/cE2rsCME5MFnZ3WV7I0WQRDBwBLgWG2B7XuioopZwp5ehp7p+ylqZhIbwz9sLdVAd3bMBXd3TnUGY213+wnG0HMs95fX1KBu/+soPhMY24tnOjigjd/dZPK93YiuJE9oHTR7yvbpG8zKpXNHbSKgx1W1t1Cy8d1e7osqq3F/O4w9XBKVVep3PyGP9lPLn5BXx6Sxw1gi++CnCP5mF8c28vCoxh1EfLWZ18FLDWw35kRgL1QoN4cVjHigjd/Y4kwdI3rak9wlqU/TzeOPNqQQFsmQfNepW/XlFIBFpcCjv+z5qQ0cs4OoL7cxGZVPTh6uCUKo+CAsOjM9ez/WAm/x3XhebhjvX2btugBrPvu4S61YO46dNV/LT5IP9csIVdh0/x5qjo86bs8Em5Z2DmreAfAMPeL9+5CusWu39zTmwVYe9yaw2K6LHOPW/n0dZEjNt+cO55K4CjzVDfA/Ntj5+BGlg9o5TyWO/9spMFiQd4+qp29G9dunWsI+qEMOu+S2jbsAb3fBXPlyv2cGefKM+fv8lZfnwaDm6EER9DzSblO1fhPFF7fre+sXuDhKkQGGp1F3am5gOsCRYTpjr3vBXA0Wao2XaPKcANgC6nqjzWj4kH+M+i7VwX25i7+pZtEFSdaoFMG9+DQe3qExNRi8ev9IGpth2xcRas+Rx6T4DWVzrnnN5Ut8g+CZvmQIfhEFjyFPOl4udvdRbYuciamNGLOHpnUVQroF5JO4nIYBHZJiI7ReSpYl6/V0Q2ikiCiCwTkfa27ZEicsa2PUFEPipjnKoS2nrgBH+dmUB0RC1eva5TqUZmFxUSWIWJt8Txv/svITigDL2BvM3hHTBvAkT0hEufc955valusWUu5J4qec2OsooeZ3Ua2DDDNed3EUdrFpkicqLwAczDWuPiYsf4A+8DVwHtgbGFycDOVGNMJ2NMDPA68Jbda0nGmBjb415H35Cq3I6eymH8l/FUD6rCxJu7Ou0DvjwJx2vknoFvbrNWhBs5yapXOEvtZlCzqXeMt0iYCnWaQ9Oerjl/3ZYQ0cO6jhd1J3a0GSrUGFPD7tHaGDO7hMO6AzuNMbuMMTnAdGBYkfOesHtaDfCe35zyOBtTjzPqo+UcPJHNxFviqF+jEoyDcKYFT8LBRLhuItRs7PzzR/bx/LrF0d1WQosZ59pR1jE3Wk1yaWtddw0nc/TOYoSI1LR7XktEhpdwWGMgxe55qm1b0XM/ICJJWHcWD9m9FCUi60TkVxEpdtUREblbROJFJD49Pd2Rt6J8UF5+Ae/+vIMRH/zOqex8vritGzERF1kfWp1vw0xYOxn6PAKtLnfNNc7WLba45vzOsH46INB5jGuv02E4VKkKCVNcex0ncrRm8XdjzNlV6I0xGcDfnRGAMeZ9Y0wLrGatZ22b9wNNjTGxwF+BqSJSo5hjJxpj4owxceHhpevtolwnv6DibhB3pZ9k5EcreOun7VzTuSELH+5XeXosOUv6dpj3MDTtBQOfLXn/svL0ukVBAayfCs37Q60I114ruKbV0ypxljVRoxdwNFkUt19JkxCmAfa/8Sa2bRcyHRgOYIzJNsYcsf28BkgCWjsYq3Kj/61LpdMLC3ly1gZO5+S57DrGGL5akczV7y5l9+FTvDc2lnfGxFIzpBKMgXCmnNPwza3WwLORk8Df0VULyqB2M6jlwXWLPb9Dxl7XFbaLihlnTdDoJWMuHE0W8SLyloi0sD3eAtaUcMxqoJWIRIlIIDAGmGu/g4jYrxhzDbDDtj3cViBHRJpj9b7SZVw9WH6B4bUFW3lkxnoa1gxm5poUrn13GYlpx0s+uJQOHM/i1s9X89x3m+geFcbCh/sxJLqSTL/hbAuegEObrTpFjQr4HUb2hWQPrVskTIGgGtC2gmYyiupvTdDoJU1RjiaLvwA5wAysO4As4IGLHWCMyQMeBBYCW4CZxphNIvKSiAy17fagiGwSkQSs5qbCNTL6ARts22cB9xpjjpbifakKlJmVy91fxvPRr0mM69GUBRP6MfWunpzOyWfEB78z8bckCpzUNDVv/T6ufPs3/th9hJeHdWDy7d0qx4R+rrB+Oqz7Cvo+Ci0HVcw1I/vAmaOeV7fIzoTN31mLOwWGVMw1/fwgZiwk/QIn9lXMNctBjBd13bqYuLg4Ex8f7+4wKp29R05z5+TV7Dp8ir8Pac/NPZud7WaacTqHp2Zv5MdNB+jTsi5v3hBd5h5Kx0/n8tx3icxdv4/oiFr854Zoh6fvUMVI3wYTB0CjWLhlrmubn+xl7IW3O8FVr0OPeyrmmo5Y9zV89wDc8X/QtEfFXfdIErzXBQa9YHUucAMRWWOMKXGQtUPJQkR+AkbZCtuISG1gujHGScM7y0+TRcVbnnSY+6esxRj44MYu9C6msGyMYcbqFF6ct5ngAD9eHxnN5e3rO3T+/ALDmj3HWLjpAN8l7CPjdA4PXdaK+we0oIqDy5h6rORl1rfJzjdU/LVzTsMnl8KpdLh3GdRoWLHXf7szVAmCZr1L3jewmjWSvHqJY4DL5/Or4eRBeDC+4hcmmjTY6iX2wB9uWRTJ0WTh6NeJuoWJAsAYc0xEXPzXU57sq5V7eHHuJpqFhfDZrd2IrFv8tAgiwpjuTYmLrMOE6esY/2U8N/dsxt+uaVfsgLmcvAKWJx1m4aaD/LT5AIdP5hDo70efVnV5eFArOjfxgS6xh3fA1NGQc9JqI28zuGKv/8PjVh//m2ZXfKIA6HILrPoYts4ved8zR2FfAtzynevufo7usorblz3vnhXsYm6EuQ9C2hpo4rmzKDn62y8QkabGmL1gTceBDqCrlHLzC3hp3ma+WrmHgW3CeWdsbInTfgO0rFedb++/hH8v3MYnS3ezctcR3h0bS7uGNTidk8dv29P5MfEAP289RGZWHiGB/gxsW4/BHRowoE04oQ5cwyvYj5Ku3xHm3Av3LHV9V81CCVMh4Wvo9zi0vKxirllUv8eshyMSplm/o19fg0td1K03YRoVMrbiQjoMtzoarPvaJ5LF34BlIvIrIEBf4G6XRaU80rFTOdw/ZS0rdh3h7n7NeXJw2wsuJFScoCr+/O2a9vRtFc6j36xn2Pu/06t5GKt2HyErt4BaIQEM7tCAwR0b0LtlXd+ci6lwlPSNs6FOFHzcH2bdDrcvcO70GsU5tBXmP2r1SBrwtGuv5SwxY2HPMvjt39Y4EGcnuIICa4GnFgNdM2rdEUGh0G4oJH4Lg/8JAVXdE0cJHJ3u40esWWa3AdOAR4Hi151UPmnnoZMMe/931uw5xr9HRfPM1e1KlSjs9Wsdzo8T+nJpm3rsOnyS0XERTB3fg/i/DeKNUdFc1q6+byaKs6Ok/wqtBlkLCg17D1JXw6IXXHvtnFPWeIrAanD9p2VbItVdrnoD6rWDb++2lnl1puSlcDyl4sZWXEjMOMg+7ljTnJs4dGchIncBE7AG1iUAPYEVWMusKh+Xm1/AA1PWcio7j2l396Rrs9rlPmdY9SA+urmrE6LzEmdHSV8CA//25/YOI6xxByv+axV8217tmuvPf8zqAXXz/yC0gWuu4SqBITBqstV7a/adzu29lTAVgmpC22ucc76yiuxrTbSYMBU6jXRvLBfgaJeSCUA3YI8xZiAQC2Rc/BDlKyYvT2bbwUxeva6TUxJFpXN2lHRVGPnZ+R90V74CDaOttvlje5x//XVTrGks+j9hNbd4o/DWcO1/rEL0kledc86sE9bYio7Xub/px37MxfGLTXThPo4miyxjTBaAiAQZY7YClWQlmMrtwPEs/vPTdga2CecKB7u8qiIWPA6Htlx4lHSVIBj1hTVd9azbIS/Hedc+uPnPOkX/i64q4PmiR1s9qZa+aS0eVF6b50DeGfc3QRWKHgMY2DDd3ZEUy9FkkSoitYA5wE8i8h3ggq9AytO8PH8zuQWGF4Z2qBxrOjhbwjSrl0vfRy9enK3THIb91+o+6az6RfZJ644mKBSu/8y76hQXctXrUK+DVb8o7zfwhKkQ1spzeiDVaW41RXroOheOFrhHGGMyjDEvAM8Bn2Gb9E/5rmU7DjN/w34eGNCSZmFOXl6yMji0Feb/FZr1caz3Ufth0P0eWPk+bPm+fNc2xrqjOLzDKmiH+shdYUBVuGGyNVPr7Dshv4yTVR5Jgr0rIPZG94ytuJCYG+HITqvTg4cp9TBYY8yvxpi5tgWNlI/Kzsvn+e8SaRYWwj39m7s7HO9T2PsoIMT6sHa0IHvFy9YUHN/dD8eSy379dV9bzRkDnrKm3PYldVvBkHesD/vFr5TtHOungfhB59HOja282g+DgGoeObmgl8+ZoFzl06W72XX4FC8O7eCb3Vhd7YfHrd5H139SulHSVYJg5OfWkNdvyli/OLgJfnjMmtW03+OlP94bdB4FXW+DZW/Bjp9Kd2xBvtU82OLSiplptzSCqlsJI/Fbq2OEB9Fkoc6TcvQ07/2ywzZ6Wmd1KbV1U6xvhv0etz6QSqtOFAx/H/athZ+eL92x2SetEeLBNb1vPEVpDX4N6ncqff1i929wItUa2+CJYsZB9gmPG3OhyUKd58V5m/ET4fkh7d0divc5tMVulPRTZT9PuyHQ4z5Y9SFsnlvy/mCrU/zVavO+/lPXT77nbgFVrV5k+Tkw6w7Iz3XsuISpVjJt4+axFRfSrDfUauZxTVEVNC+x8kgF+bD4VeubTFgLAH7ecpBFWw7y1FVtaVTLM6cd8Fg5p2DmrVZTgjO+1V/+EqSssqbOduRbZnYmbJtvDfqL6le+a3uLui2t+sXsO+GrEVDDgSk7tsy1CskBHroOip+f9f/kktdg9nirtlKSOlHl+3LiAE0WlVnqalj6b+uDaPzPZEkwL8zbRMt61bmjd5S7o/MuZ3sfbXfeKOkqgTDqc5h1p1XMdUTMjVY33cqk00ir19f6adZ6GSWp1RS6j3d9XOURe7N1R5myyrH9c066Nh40WVRuhWshp2+FH57gg5CHSDl6hmnjexJYRVsoSyVhivVh1f8p546Srh0J43923vl81cCnrYevqNkY7l/u7ijOocmiMkteZhUI2wyG397gYF4ow2NG06tFmLsj8y4HN1tzL0X1s6bUUMoH6dfHyiovB/augsg+mP5PsTkomhf8P+O5Hh40QMkb2I+Svs7Hex+pSk2TRWWVtsaaFyeyDws2p3Pr8bshqDphP9xjFWpVyYr2PvKVUdJKFUOTRWWVvAwQTjXswUvzNhPesBmBoz6zBpLNd3AVs8pu3VewYYZVp/C1UdJKFaHJorJKXgr1O/Lu8sMcOJHFy8M74t/qUqvNff1Ua2CZurCDm6xR2s0HOL5EqFJeTJNFZZSXDSl/kFqrK58t283ouIg/16no/6Q1oGz+o1bhVp0vO9MaTxFcE677ROsUqlLQZFEZpa2FvDO8urkureqH8sw17f58zc/fms46KNQq3Ga7vv+2VzEGvn8EjiZZvydfHyWtlI0mi0ro4IZFFBghrUYMX93ZnZpVA87dIbS+VbA9vMO6w/DAufXdZu1k2PgNDHgGovq6OxqlKowmi0pm56GT7FnzIzv9Ivlg/CDqVg8qfsfm/a3pAzZMt6a7VnBgI/zwBDQfCH3/6u5olKpQLk0WIjJYRLaJyE4ROW/iEhG5V0Q2ikiCiCwTkfZ2rz1tO26biFzpyjgri9Rjp7nj02V0Zjv1O19G45Lmfur3uDXN9Q+PWQXdyiw705rNtWptrVOoSsllyUJE/IH3gauA9sBY+2RgM9UY08kYEwO8DrxlO7Y9MAboAAwGPrCdT5XRocwsbvp0Fc1zthJMDjXbOTB1tp+/1RwVXNMq6FbW+oUxMO9hOLoLRn4G1cPdHZFSFc6VdxbdgZ3GmF22VfWmA8PsdzDGnLB7Wg1ryRds+003xmQbY3YDO23nU2WQcTqHWz77g0OZ2bwSexwQaHaJYwdXr2cljKNJVmHXHfWL7JPwf8/C4Z0Vf22w5n1KnGXN5hrZxz0xKOVmrkwWjYEUu+eptm3nEJEHRCQJ687ioVIee7eIxItIfHp6utMC9yWnsvO47fPV7Eo/xcSb42icEQ8NOlnNKY6K6metIb1xJqz90nXBFqdwlPTy92D6OPeMLl8z2fqd9dE6haq83F7gNsa8b4xpATwJPFvKYycaY+KMMXHh4do0UFRWbj7jv4xnY9px3hsXS5+oUEj5wxpHUVp9H7UGoC14Ag4kOjvUC1v7pTVKuv0wa/rviu6dlX3SWrGu1RXWOgNKVVKu/K8/DYiwe97Etu1CpgPDy3isKiI3v4C/TFvH8qQjvDGyM1d2aACp8ZCXVbamFD9/a6K84Fq28ReZzg+6qAOJVnJqPtBal3rAU9Y04BW5gljKSijI0+YnVem5MlmsBlqJSJSIBGIVrM9ZH1JEWtk9vQbYYft5LjBGRIJEJApoBfzhwlh9SkGB4YlZG/hp80FeGtaB67o0sV6wzQdFs15lO3H1cKvAe3SXVfB15Tf87EwrKQXX+rP3Ub/HrSax+Y9V3Ojy5GXgVwUielTM9ZTyUC5bz8IYkyciDwILAX9gkjFmk4i8BMQbY+YCD4rIICAXOAbcajt2k4jMBDYDecADxph8V8XqLZbtOMyuwyc5mZ3Hyaw861+7n09l55GZnceJM7kcPpnD41e24ZZekX+eIHlp6esVRUX2gYHPwC//sH6Ou73c7+s89r2Pbp33Z++jwrubj/pYiWT8YmsJU1dKXgaNu0JgNddeRykP59LFj4wxPwA/FNn2vN3PEy5y7CvAK66LzrvM37CfB6auPfu8ip9QPbgK1YP+fNSuFkiTOiGEBlUhJqIWo7vZteTlZlnLqMbdWf5g+jwKe5bDgiehSZyVgJxpzRdW76NLnz2/+Se0vnV38+Uwq/A94mMQF63BkX3Smhqlz8OuOb9SXkRXyvMCR0/l8Px3iXRuUpPPbu1GaHAVgqr4IaX5kExbY9UrnDFFhZ8fjJgIH/e1xl/cvQSCa5T/vGCNkl7wJLS41EpKxYnqZ00LvuRVK5l0ucU51y4qZSWYfK1XKIUH9IZSJXth7iZOZOXyxshowkODCA7wL12igD/rFU3LWK8oqnq4NZHesd0wb4Jz6hdZJ6zkE1LHSkYX633U7zGrd9YPj7tudHnyMvAL0HqFUmiy8Hj/t+kAc9fv4y+XtqJNg9Cynyh5KTTsDFVrOS+4yN5WU9GmbyF+UvnOZYyVdI7tts3mWkJXaD9/q/B9dnS5C3pn7V6q9QqlbDRZeLCM0zn8bU4i7RrW4L4BLcp+otysso+vKEnvR6DlIPjxadi/vuzniZ9kJZ2Bf7OSkCOq17MSiytGl2dnwr512gSllI0mCw/20vebOXYqhzdGdibAvxx/qrR4yM92zQefn59VZA4JsybayzpR4iHn2b/eSjYtLiv9KOmovtZ04Ru/saYPd5a9q7ReoZQdTRYe6petB/l2bRr3DWhBx8Y1y3cyZ9criqpWF0ZOgmN7YN5DpfuGf7ZOEQbXlVCnuJC+j1oD9354wiqQO0PyUlu9QqckUwo0WXikE1m5PPNtIq3rV+fBS1uW/4TJy5xfryiqWS+47DnY9D9Y/aljxxgDc/8CGXutZFOtbtmu7edn1S9C6lh3N86oX+j4CqXOoV1nPdCr87dwKDOLj2/uTVCVcs7MXliv6D7eOcFdzCUTrPEXC5+BbQtKHv+Qcxr2LofL/l72UeWFCntnTb7W6iE14qOyn6uwXtHnkfLFpJQP0TsLD7N0RzrTV6cwvl9zoiOccCeQutpWr6iAJUD9/GD4R9DycsjKgDPHLv7Iz4Ye90JvJw16i+wNvR6wJh48sa/s59F6hVLn0TsLD3IyO4+nZm+keXg1HhnU2jknTV4G4gdNezrnfCWpFgZjp1bMtYrT9XZrOvP108u+9Gnybzq+Qqki9M7Cg7y2YAv7jp/hjZGdCQ5w0sKAycuggYvrFZ4krIVVyE+YWvautMnLrGlMAkOcG5tSXkyThYdYnnSYr1fu5Y7eUXRtVsc5Jy2cD6qyNafEjIMjO6wp2Usr6wTsS6h8vzOlSqDJwgUyTufwwtxN/OvHrfyYeICDJ7Iuuv/pHKv5KTIshMeuaOO8QCqyXuFJOoyAgJCyrXuRovUKpYqjNQsn25dxhlsm/UHy4VOIQG6+1RTSoEYwMRG1iI6oRUxELTo1qUn1IOvX/8bCbew9epoZd/ekaqCTmp/AGisgfuXvaeRtgkKh3VBI/BYG/xMCqjp+bOH4iiY6vkIpe5osnGj7wUxunfQHJ7Py+PquHsRE1GLz/hMk7M1gfWoGCSkZ/LjpAAB+Aq3qhdK2YShz1+/jll7N6NE8zLkBJS+DhtHW/EmVTcw42DAdts6HTiMdP07rFUoVS5OFk6zZc5Q7vognsIofM+7pRftG1pTdXZrWpkvTPxcbOnYq52ziWJ+SwW/b04kKq8aTg9s6N6DcM1YzVI97nHtebxHZF2o2tZqiHE0WhfWKsvaiUsqHabJwgp82H+TBqWtpVKsqX97RnYg6F/5WWrtaIAPa1GNAm3oAGFuPnVJPOV6S1NWQn1P56hWF/PwgZiz8+jocT4OajUs+Zm/h+hWV9Hem1EVogbucZqzeyz1fxdO2QSiz7u110URRHBFxfqKAih9f4YmixwDGao5yRPJS8A+EJt1cGpZS3kiTRRkZY/jvLzt4cvZG+rQKZ+r4noRVD3J3WH+qzPWKQnWaQ7PesG6KY2MukpdBY61XKFUcTRZlkF9geGHuJv79f9sZEduYT2+Jo1qQB7XoFdYrtDnFKnQfTbLmx7qYrBOwX8dXKHUhmixKKTsvn4emrWPyij2M7xvFm6OiCaziYb/GlD8qd73CXvvhEFCt5DEXe1eCKdBkodQFeNDXYc93IiuXe79aw/KkIzxzdVvu7leO1esKbZkHKz8s/3nsZe7XekWhoOrQfpg1dfrg1y7cxKT1CqUuSpOFA87k5PP1yj189GsSx8/k8tYN0VzXpUn5T2wM/PyyNQNruBNHbtdobH2jDq7hvHN6s5hxsH4qbP0eOt9Q/D5ar1DqojRZXERWbmGS2MXhk9n0bVWXR69oQ4wzpg4HSFsDh7fB0Pegyy3OOac6X7PeUMs25qK4ZJF13KpX9Hu84mNTyktosihGVm4+0/7Yy4dLkjiUmc0lLcL48KYudIt00gR/hRKmQJWq1l2Ach0/P4i5EZa8BhkpUCvi3Ne1XqFUiTRZ2MnOy2fG6hQ+WJzEgRNZ9Iiqw7tjY+np7Gk4wJoRduNsaD9Um4sqQvQYWPJPa8xF0TsIrVcoVSKXJgsRGQy8A/gDnxpjXivy+l+Bu4A8IB24wxizx/ZaPrDRtuteY8xQV8WZk1fAzPgU3l+8k/3Hs+gWWZu3RkdzSYsyrgntiG3zIfu49Y1XuV7tSKt3WMJU6PvYuUu+Ji+zEkVpJhxUqpJxWbIQEX/gfeByIBVYLSJzjTGb7XZbB8QZY06LyH3A68Bo22tnjDExroqv0J4jpxj3ySrSMs7QpWkt3hgZTe+WYa4ZVW1v3RSoGaHdWytSzDiYc5/V7FQ4E2/Wcdi/XusVSpXAlQMEugM7jTG7jDE5wHRgmP0OxpjFxpjTtqcrASd0MSqdJrVDiIuszeQ7ujP7vkvo06qu6xPFiX2wazFEj7Xa01XFaD8MAqufO+ZC6xVKOcSVn1SNgRS756m2bRdyJ7DA7nmwiMSLyEoRKbYCLCJ32/aJT09PL1OQ/n7CO2Ni6d863PVJotD66dYHVMzYirmesgRWszoTbJoDOaesbVqvUMohHvG1VkRuAuKAN+w2NzPGxAHjgLdF5LwRcMaYicaYOGNMXHh4eAVFW07GWO3mzXpbcxepihUzDnIyYcv31vPdS62FjrReodRFuTJZpAH2fRSb2LadQ0QGAX8Dhhpjsgu3G2PSbP/uApYAsS6MteKkrrbWh44Z5+5IKqemvaxid8IUOJMBBzZoE5RSDnBlslgNtBKRKBEJBMYAc+13EJFY4GOsRHHIbnttEQmy/VwX6A3YF8a9V8IUa33o9sNK3lc5n58fRI+D3b/Bxm+0XqGUg1yWLIwxecCDwEJgCzDTGLNJRF4SkcJusG8A1YFvRCRBRAqTSTsgXkTWA4uB14r0ovJOuWesdaHbD7PWiVbuETMWMPDzS+AfpPUKpRzg0nEWxpgfgB+KbHve7udBFzhuOdDJlbG5xdb5kH1Cm6DcrVZTiOpn3V006wMBwe6OSCmPpyO4zxyDr0Y4vn/DaLj6TfAvw68uYYr1QdVMmz3cLuZGK1loE5RSDtFkIX5QzcGeVHlZsOYLqFYPLv1b6a5zPBWSFkP/J3VshSdoP8xa90Pv8pRyiCaL4Jpw4zeO7z/nAfjtDWutiJaXOX7c+umAsa0LrdwuoCpc+5a7o1DKa+hX3NK6+g0Ibwvf3g0n9jt2zNmxFX2gTpRr41NKKRfQZFFagSFww2TIPQ2z74T8vJKPSfnDWgc6VicNVEp5J00WZRHeBq79D+z53Zr2uiQJX1vrQLdz2cS5SinlUposyip6DMTeDEvfhJ2LLrxfzmlI/B90GG6tB62UUl5Ik0V5XPU61Gtnq1/sK36frd9bcxFprxullBfTZFEegSEwarK16t2sC9QvEqZYcxE1vaTCw1NKKWfRZFFe4a1hyNuwdzksfuXc1zJSYNev1lxEOrZCKeXF9BPMGTrfAF1uhWVvwQ67+oWOrVBK+QhNFs5y1b+gXgf4djwcT7ONrZhiLZtau5m7o1NKqXLRZOEsAVWt8Rf5OaK4lGQAAAZZSURBVDDrDkheBsd2Q+xN7o5MKaXKTZOFM9VtBUPegZSVMPMWa73ndkPcHZVSSpWbJgtn6zQSut4GZ45aYysCq7k7IqWUKjedSNAVBr8GIXWhyy3ujkQppZxCk4UrBFSFy55zdxRKKeU02gyllFKqRJoslFJKlUiThVJKqRJpslBKKVUiTRZKKaVKpMlCKaVUiTRZKKWUKpEmC6WUUiUSY4y7Y3AKEUkH9pTjFHWBw04KxxP42vsB33tPvvZ+wPfek6+9Hzj/PTUzxoSXdJDPJIvyEpF4Y0ycu+NwFl97P+B778nX3g/43nvytfcDZX9P2gyllFKqRJoslFJKlUiTxZ8mujsAJ/O19wO+95587f2A770nX3s/UMb3pDULpZRSJdI7C6WUUiXSZKGUUqpElT5ZiMhgEdkmIjtF5Cl3x+MMIpIsIhtFJEFE4t0dT2mJyCQROSQiiXbb6ojITyKyw/ZvbXfGWFoXeE8viEia7e+UICJXuzPG0hCRCBFZLCKbRWSTiEywbffKv9NF3o83/42CReQPEVlve08v2rZHicgq22feDBEJdOh8lblmISL+wHbgciAVWA2MNcZsdmtg5SQiyUCcMcYrBxOJSD/gJPClMaajbdvrwFFjzGu2pF7bGPOkO+MsjQu8pxeAk8aYf7sztrIQkYZAQ2PMWhEJBdYAw4Hb8MK/00Xezw14799IgGrGmJMiEgAsAyYAfwW+NcZMF5GPgPXGmA9LOl9lv7PoDuw0xuwyxuQA04Fhbo6p0jPG/AYcLbJ5GDDZ9vNkrP+RvcYF3pPXMsbsN8astf2cCWwBGuOlf6eLvB+vZSwnbU8DbA8DXArMsm13+G9U2ZNFYyDF7nkqXv4fiI0B/k9E1ojI3e4OxknqG2P2234+ANR3ZzBO9KCIbLA1U3lFk01RIhIJxAKr8IG/U5H3A178NxIRfxFJAA4BPwFJQIYxJs+2i8OfeZU9WfiqPsaYLsBVwAO2JhCfYay2U19oP/0QaAHEAPuBN90bTumJSHVgNvCwMeaE/Wve+Hcq5v149d/IGJNvjIkBmmC1pLQt67kqe7JIAyLsnjexbfNqxpg027+HgP9h/Ufi7Q7a2pUL25cPuTmecjPGHLT9z1wAfIKX/Z1s7eCzgSnGmG9tm73271Tc+/H2v1EhY0wGsBjoBdQSkSq2lxz+zKvsyWI10Or/27ufEBujMI7j3x+iMYoUG4WGDQqlLAw1payp8Z8mKwsbOxEpZYuNMgtqZPzPMFmymFgIochYyWI2bKQoEo/FeUZDzbzXDO7c5vdZ3Tn3zOmcTvc+9z3nfZ+TdwdMBbYBvXXu05hIas4NOiQ1AxuAFyP/V0PoBTrydQdwq459+SsGv1TTJhponnLz9CzQHxEnhrzVkPM03HgafI7mSJqVr5soN/L0U4JGe1areY4m9N1QAHkr3ClgMnAuIo7XuUtjIqmFcjUBMAW42GhjknQJaKOkUn4LHAVuAleB+ZRU9FsiomE2jIcZUxtleSOAN8DeIev945qktcA94DnwPYsPUdb5G26eRhjPdhp3jpZTNrAnUy4MrkbEsfyOuAzMBp4CuyLiS2V7Ez1YmJlZtYm+DGVmZjVwsDAzs0oOFmZmVsnBwszMKjlYmJlZJQcLs3FAUpuk2/Xuh9lwHCzMzKySg4XZH5C0K88IeCapMxO1fZR0Ms8MuCtpTtZdKelBJqHrGUxCJ2mxpDt5zsATSYuy+RmSrkt6Jak7nyo2GxccLMxqJGkJsBVozeRs34CdQDPwOCKWAX2Up7MBzgMHImI55cngwfJu4HRErADWUBLUQcl0uh9YCrQArf98UGY1mlJdxczSemAV8Ch/9DdREuV9B65knQvADUkzgVkR0ZflXcC1zNs1LyJ6ACLiM0C29zAiBvLvZ8BCyoE1ZnXnYGFWOwFdEXHwl0LpyG/1RptDZ2h+nm/482njiJehzGp3F2iXNBd+nje9gPI5GsziuQO4HxEfgPeS1mX5bqAvT2EbkLQx25gmafp/HYXZKPiXi1mNIuKlpMOUUwgnAV+BfcAnYHW+946yrwEl/fOZDAavgT1ZvhvolHQs29j8H4dhNirOOms2RpI+RsSMevfD7F/yMpSZmVXylYWZmVXylYWZmVVysDAzs0oOFmZmVsnBwszMKjlYmJlZpR+As2aDhlFxegAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe5990fe5c0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEWCAYAAAB1xKBvAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvNQv5yAAAIABJREFUeJzt3XuYXXV97/H3J2EgEwhMTMJtAibSNgRITMrA0QYV5QiIVSIKEZHSHoX6HFovx/IYeixy0JZYeo6aVsS05kF6KJgCyaGiIhgwVUCZXOQiIEJDmQGTkBuJScgk+Z4/9hqyZ7JnZu3Ze83al8/refLM3r912b+VPbO/e/0u358iAjMzs6GMyrsCZmZWHxwwzMwsFQcMMzNLxQHDzMxSccAwM7NUHDDMzCwVBwyzKpB0s6Qvpdx3raT/Wul5zEaaA4aZmaXigGFmZqk4YFjTSJqCrpL0mKTfSvqWpKMkfV/SNkn3SxpftP/7JT0paYukByVNL9o2W9Kq5LjvAGP6vdYfSlqTHPuQpJnDrPPlkn4taZOkuyUdm5RL0lckrZf0qqTHJZ2SbDtP0i+TunVL+oth/YeZ9eOAYc3mg8C7gd8D3gd8H/hLYBKFv4dPAkj6PeA24NPJtu8B/ybpYEkHA8uAfwbeAPxrcl6SY2cDi4E/BSYA3wTulnRIORWV9C7geuAi4BjgBeD2ZPPZwNuT6zgi2Wdjsu1bwJ9GxDjgFGB5Oa9rNhAHDGs2fx8R6yKiG/h34GcRsToidgFLgdnJfvOAeyLivojoAf4OaAX+AHgL0AJ8NSJ6IuIO4NGi17gC+GZE/Cwi9kbEt4HXkuPKcQmwOCJWRcRrwNXAWyVNAXqAccCJgCLiqYh4OTmuBzhJ0uERsTkiVpX5umYlOWBYs1lX9HhnieeHJY+PpfCNHoCI2Ae8CLQn27qjb+bOF4oevxH4bNIctUXSFuC45Lhy9K/Ddgp3Ee0RsRz4B+DrwHpJiyQdnuz6QeA84AVJP5b01jJf16wkBwyz0l6i8MEPFPoMKHzodwMvA+1JWa/jix6/CPx1RLQV/RsbEbdVWIdDKTRxdQNExMKIOBU4iULT1FVJ+aMRcT5wJIWmsyVlvq5ZSQ4YZqUtAd4r6SxJLcBnKTQrPQQ8DOwBPimpRdIFwOlFx/4j8AlJ/yXpnD5U0nsljSuzDrcBfyJpVtL/8TcUmtDWSjotOX8L8FtgF7Av6WO5RNIRSVPaq8C+Cv4fzF7ngGFWQkQ8A3wU+HvgFQod5O+LiN0RsRu4APhjYBOF/o67io7tBC6n0GS0Gfh1sm+5dbgf+CvgTgp3NScAH042H04hMG2m0Gy1Ebgh2XYpsFbSq8AnKPSFmFVMXkDJzMzS8B2GmZml4oBhZmapOGCYmVkqDhhmZpbKQXlXoJomTpwYU6ZMybsaZmZ1Y+XKla9ExKQ0+zZUwJgyZQqdnZ15V8PMrG5IemHovQrcJGVmZqk4YJiZWSoOGGZmlkpD9WGU0tPTQ1dXF7t27cq7KpkaM2YMkydPpqWlJe+qmFmDaviA0dXVxbhx45gyZQp9k4s2johg48aNdHV1MXXq1LyrY2YNquGbpHbt2sWECRMaNlgASGLChAkNfxdlVpMeWwJfOQWubSv8fKxxs8k3/B0G0NDBolczXKNZzXlsCfzbJ6FnZ+H51hcLzwFmXpRfvTKS2R2GpMXJAvVPDLHfaZL2SPpQUdleSWuSf3dnVUczs4r86Lr9waJXz85CeQPKsknqZuDcwXaQNBr4MvDDfpt2RsSs5N/7M6rfiNiyZQs33nhj2cedd955bNmyJYMamVnVbO0qr7zOZRYwImIFhcVlBvPnFBaHWZ9VPcq1bHU3cxYsZ+r8e5izYDnLVndXdL6BAsaePXsGPe573/sebW1tFb22mWXsiMnllde53Dq9JbUDHwC+UWLzGEmdkh6RNHeI81yR7Nu5YcOGiuq0bHU3V9/1ON1bdhJA95adXH3X4xUFjfnz5/Pcc88xa9YsTjvtNN72trfx/ve/n5NOOgmAuXPncuqpp3LyySezaNGi14+bMmUKr7zyCmvXrmX69OlcfvnlnHzyyZx99tns3LlzoJeztJqoo9IydNY10NLat6yltVDegPIcJfVV4HMRUWq94TdGRAfwEeCrkk4Y6CQRsSgiOiKiY9KkVPmzBnTDvc+ws2dvn7KdPXu54d5nhn3OBQsWcMIJJ7BmzRpuuOEGVq1axde+9jV+9atfAbB48WJWrlxJZ2cnCxcuZOPGjQec49lnn+XKK6/kySefpK2tjTvvvHPY9TH2d1RufRGI/R2VDhpWrpkXwfsWwhHHASr8fN/ChuzwhnxHSXUAtyejeyYC50naExHLIqIbICKel/QgMBt4LotKbN6xm3Vbd7F77z5e2lL6m/tA5cNx+umn95krsXDhQpYuXQrAiy++yLPPPsuECRP6HDN16lRmzZoFwKmnnsratWurVp+mNFhHZYP+oVuGZl7UNL83uQWMiHj9U1PSzcB3I2KZpPHAjoh4TdJEYA7wt1nUYfOO3XRv3sm+ZF3zieMOYcO21w7Y79i21gPKhuvQQw99/fGDDz7I/fffz8MPP8zYsWM588wzS86lOOSQQ15/PHr0aDdJVSrvjsrHlhSC09auQlv3Wdc0zQeO1bcsh9XeBjwMTJPUJeljkj4h6RNDHDod6JT0C+ABYEFE/DKLOq7buuv1YAFw6VuO55CD+v6XtLaM5qpzpg37NcaNG8e2bdtKbtu6dSvjx49n7NixPP300zzyyCPDfh0rQ54dlW4OszqW2R1GRFxcxr5/XPT4IWBGFnXqb/fevt0nZ047EoB/fuQ/eWXbaxzb1spV50xj7uz2Yb/GhAkTmDNnDqeccgqtra0cddRRr28799xzuemmm5g+fTrTpk3jLW95y7Bfx8pw1jV9J1vByHVUujnM6pii6Bt2vevo6Ij+Cyg99dRTTJ8+veT+T7/86gFBA+Dg0aM48ZjDM6ljlga7Vusnr2aha9uAUn9zgms978ZGnqSVySCjITVFapCBHHXEmD59GACjJI46YkyOtbIRkVdH5RGTk+aoEuXWl/t6ak7DJx8czPixB9M+vpWDRxf+Gw4ePYr28a2MH3twzjWzhtVk4/aHzX09Namp7zCgEDQcIGzE9H5D9jfnwbmvpyY1fcAwG3FNNG5/2PIe+mwlNXWTlJnVqCbL0VQvHDDMrPa4r6cmOWDUmMMOOyzvKpjlr8lyNNUL92H056F8ZrXBfT01xwGjWAbLLc6fP5/jjjuOK6+8EoBrr72Wgw46iAceeIDNmzfT09PDl770Jc4///xqXIGZWWbcJFUsg+UW582bx5Il+8eOL1myhMsuu4ylS5eyatUqHnjgAT772c/SSDPuzawx+Q6jWAZD+WbPns369et56aWX2LBhA+PHj+foo4/mM5/5DCtWrGDUqFF0d3ezbt06jj766GG/jplZ1hwwimWUtuHCCy/kjjvu4De/+Q3z5s3j1ltvZcOGDaxcuZKWlhamTJlSMq25lcn9T2aZcpNUsYyG8s2bN4/bb7+dO+64gwsvvJCtW7dy5JFH0tLSwgMPPMALL7xQ0fkzV85ypnktfVoLqSS87Ks1OAeMYhkN5Tv55JPZtm0b7e3tHHPMMVxyySV0dnYyY8YMbrnlFk488cTq1B+q/6FVzgdxnh/aGfQ/laUWApZZxpo6vXmjeeoXnUz/7h8euM5DJUHvK6cM0Ex3HHzmieHvW215pw3P89rBzXE2bOWkN/cdRiPZtbX637LLGQiQZ/6fvFNJ5HntvruxEeKAkacdm2Ddk/DS6sLPHZsqO9++PaXLK/nQKueDOM8P7bxTSeR57Xk3x1nTaIqAUZPNbjs2Fb4J7t1deL53d+H5MINGRMCo0aU3VvKhVc4HcZ4f2nmnksjz2p3ZtT5kMShihAdaNPyw2jFjxrBx40YmTJiApGGfZ/OO3azbuovde/dx8OhRHHXEmMrW0dj2MkS/5WFjX6F87BvKOlVEsHHjRsaMaS18SFVzrepy1m8oZ98s2tzzTCWR5zoXXsWv9mWQRSKTcw6h4Tu9e3p66Orqqmiew47de9iyo4d9Rf9VowRtY1sYe/AwY+6W/xx4W9vxZZ9uzJgxTJ48mZanltZ+52f/X3QYvHPeHbqDK/f/00ZeFoMiqnROr+ldpKWlhalTp1Z0jjkLltO9ZecB5e1trfx0/ruGd9KvXJjNqJp6SNhWzmpqOXyLqjtexa/2ZdFsmENTZMMHjCGl+Pb6UolgMVj5stXd3HDvM7y0ZSfHtrVy1TnTmDu7ve9OZ11T+lthM+T7L+cX3Ut1plMPXxSaWRbNhjk0RTZFp/eAUg5HPLatteThpcqXre7mJ0tv5Ds7Lue5Qz7Cd3Zczk+W3siy1d19d8y7kzZP5Ywocodu9XlG+sjLYlBEDgMtmjtgpByOeNU502ht6TsCqbVlNFedM+2AU665ZxHXaRGTR73CKMHkUa9wnRax5p5FB77+zIsKzU/Xbin8bIZgAeX9ouc9v6LReM5GPrL4gpjDl85MO70lLQb+EFgfEacMst9pwMPAhyPijqTsMuDzyS5fiohvD/V6pTq9B1XG7OBUzUxA1zUnMHnUKweW75vI5OueS1+3Rpe2I9sdutWV94x0qzm11Ol9M/APwC0D7SBpNPBl4IdFZW8AvgB0UPhEXynp7ojYXNXaldEGOHd2e8kA0d+xozamLk8bhBpS2jZ3d+hWl5v4rAKZBoyIWCFpyhC7/TlwJ3BaUdk5wH0RsQlA0n3AucBtVa1gBh3Pu1qPZuzOl0uXFz1ftrqbq+96nJ09ewHo3rKTq+96HKB5gkZa7tCtHs/ZsArk2ochqR34APCNfpvageLf6q6krNQ5rpDUKalzw4YN5VUggzbAse+5jj2jx/Qp2zN6DGPf07df5IZ7n3k9WPTa2bOXG+59ZtivbTakcvqPmrlzvJmvfRB5D6v9KvC5iNg33FnYEbEIWASFPoyyT1Dtb68zLyr8pxY1oRxUhaG6ZlWRtokvq/kv9TAJ03N/BpR3wOgAbk+CxUTgPEl7gG7gzKL9JgMPjnTlhi1FEDq2rbXkZMCBhvCaVU2aL0lZzH+plw9iz/0ZUK5NUhExNSKmRMQU4A7gv0fEMuBe4GxJ4yWNB85OyhpGOUN1Ad8i28jKonO8XrLqemDAgDK9w5B0G4U7hYmSuiiMfGoBiIibBjouIjZJ+iLwaFJ0XW8HeKPo7dhONUqqXr6ZWePIonO8Fj6I0zSJeWDAgBo++WBDyGrsfD20J1s+spj/kvfvcdprarK5P15xLws5NgnFAN/ABipPxTN+bTBZzCLOIpVFOb/HaZvEmjltzxDy7vSuDzk3Ca1jIkdz4JDhQvkwuWPPhpLBCEKgune15fwel9Mk5rk/JTlgpJHzh+v1uy/k+pZ/Yqx2v162Iw7m+p4L+Vq/fVPPHq+F9mRrPtX+IC7n99h9ExVzk1QaOX+4dh7+bub3fJyufRPZF6Jr30Tm93yczsPf3We/3tnj3Vt2EuyfPX5AplxwUj9rDOX8Hue97nsDcMBII+cP16vOmcZ9o9/BGbsX8qbXbuWM3Qu5b/Q7DhiCW9bscf/xWCMo5/fYfRMVc5NUGjkvdpR2CG5Zs8ed1M8aQbm/x+6bqIgDRho18OGaJltu2bPH/cczNA89rn3+PR4xDhhp1cEv5VXnTOuTAReGmD1ug/OESbM+3IfRQObObuf6C2bQ3taKgPa2Vq6/YIbTpQ9XvaSyMBshvsNoMGkXerIUPPTYrA/fYZgNxEOPzfpwwGhiy1Z3M2fBcqbOv4c5C5aXnq/RzDz02KwPN0k1KS8Rm0INjI4zqyUOGE1qsEl+DhhF6mB0nNlIcZNUk/ISsWY1rEYXTHPAaFIDTebzErFmOavhpQccMJpU2UvEmtnIqOH5P+7DaFJlLRFrZiOnhuf/OGA0MU/yM6tBNbxuh5ukbEier2E2gmp4/o/vMGxQnq9hNsJqeP6PA4YNqhbma6RedtasUdTo/B8HDBtU3vM1fIdjVjvch2GDynu+RlnLzppZpnyHYYMqd1GmtM1HaffL+w7HzPbL7A5D0mJJ6yU9McD28yU9JmmNpE5JZxRt25uUr5F0d1Z1tKGVsyhTb/NR95adBPubj/qPqkq7H+R/h2Nm+ykisjmx9HZgO3BLRJxSYvthwG8jIiTNBJZExInJtu0RcVi5r9nR0RGdnZ2VVt2Gac6C5SXXFG9va+Wn899V9n5wYB8GFO5wvJKgWXVIWhkRHWn2zaxJKiJWSJoyyPbtRU8PBbKJXDZi0jYfldPM1PQz0h9bUpPDK6055dqHIekDwPXAkcB7izaNkdQJ7AEWRMSyPOpn5Tm2rbXknUP/5qO0+/VKOyO94Ybf9iah680r1JuEDhw0LBe5jpKKiKVJM9Rc4ItFm96Y3CJ9BPiqpBMGOoekK5I+kM4NGzZkXGMbTNqEhlkkPiynX6Ru1HASOmtONTGsNiJWAG+SNDF53p38fB54EJg9yLGLIqIjIjomTZo0EtW1AaTtIC+nIz2thhx+W8NJ6Kw55dYkJel3gOeSTu/fBw4BNkoaD+yIiNeSADIH+Nu86mnlSdt8VO3Ehw05/LaGk9BZc8osYEi6DTgTmCipC/gC0AIQETcBHwT+SFIPsBOYlwSP6cA3Je2jcAe0ICJ+mVU9rTGU2y+SVq79Imdd07cPA2omCZ01pyxHSV08xPYvA18uUf4QMCOrelljKneCYRq5pyWp4SR01pw809saQhbDb2sh8WKtJqGz5uSAYQ2jnH6RNE1NDdkvYlaBmhglZTaS0g7BdVoSs74cMKzppB2Cm8V8EbN65iYpazppm5qaPi1JOZzCpCk4YFjTKWcIbrXnizQkpzBpGm6SsqbjpqYqcwqTpuE7DGs6eTc1NVySRKcwaRoOGNaU8mpqyn0yYBacwqRpuEnKbAQ1ZJLEs64ppCwp5hQmDckBw2wENeRkwJkXwfsWwhHHASr8fN9Cd3g3IDdJmY2grJIk5s4pTJqC7zDMRpBHaFk98x2GWZWkGf2U9wgts0o4YJhVQTmjnzwZ0OqVm6TMqqAhRz+Z9eOAYVYFDTn6yayfVAFD0qckHa6Cb0laJensrCtnVi+cCt2aQdo7jP8WEa8CZwPjgUuBBZnVyqzOePSTNYO0nd5Kfp4H/HNEPClJgx1g1kw8+smaQdqAsVLSD4GpwNWSxgH7squWWf3JYvRTwyUqtLqWNmB8DJgFPB8ROyS9AfiT7KplZg2ZqNDqWto+jLcCz0TEFkkfBT4PbM2uWmbmobpWa9IGjG8AOyS9Gfgs8BxwS2a1MjMP1bWakzZg7ImIAM4H/iEivg6My65aZuahulZr0gaMbZKupjCc9h5Jo4CW7KplZh6qa7UmbcCYB7xGYT7Gb4DJwA1DHSRpsaT1kp4YYPv5kh6TtEZSp6QzirZdJunZ5N9lKetp1jDmzm7n+gtm0N7WioD2tlauv2CGO7wtNyq0NKXYUToKOC15+vOIWJ/imLcD24FbIuKUEtsPA34bESFpJrAkIk5MRmF1Ah1AACuBUyNi82Cv19HREZ2dnamux8zMQNLKiOhIs2/a1CAXAT8HLgQuAn4m6UNDHRcRK4BNg2zfHvsj1qEUggPAOcB9EbEpCRL3AeemqauZmWUj7TyM/wmc1ntXIWkScD9wR6UVkPQB4HrgSOC9SXE7ULyqfFdSVur4K4ArAI4//vhKq2NmZgNI24cxql8T1MYyjh1URCyNiBOBucAXh3H8oojoiIiOSZMmVaNKZmZWQto7jB9Iuhe4LXk+D/heNSsSESskvUnSRKAbOLNo82TgwWq+nlkjcQoRGwmpAkZEXCXpg8CcpGhRRCyt9MUl/Q7wXNLp/fvAIRTuXu4F/kbS+GTXs4GrK309s0bkFCI2UlIv0RoRdwJ3lnNySbdRuFOYKKkL+ALJ/I2IuAn4IPBHknqAncC8pBN8k6QvAo8mp7ouIgbsPDdrZoOlEHHAsGoaNGBI2sb+kUt9NgEREYcPdnxEXDzE9i8DXx5g22Jg8WDHm5lTiNjIGTRgRITTf5jVuGPbWukuERycQsSqzWt6m9W5rFKILFvdzZwFy5k6/x7mLFjOstXdFZ3P6l/qPgwzq01ZrPbnjnQrxQHDrAFUe7U/d6RbKQ4YZk0mzZwNd6RbKe7DMGsivU1N3Vt2EuxvaurfP+G1OKwUBwyzJpJ22ddyO9LdQd4c3CRl1kTSNjWV05HuDvLm4YBh1kTKmbORtiPdHeTNw01SZk0kizkb7iBvHg4YZk0ki2Vf3UHePNwkZdZkqj1n46pzpvXpw4DqzDS32uOAYWYVyWKmudUmBwwzq1i171qsNjlgmNmI8cqA9c0Bw8xGhOdr1D+PkjKzEZF2lrnVLgcMMxsRnq9R/xwwzGxEeL5G/XPAMLMRkdXKgDZy3OltZiPC8zXqnwOGmY0Yz9eob26SMjOzVBwwzMwsFQcMMzNLJbOAIWmxpPWSnhhg+yWSHpP0uKSHJL25aNvapHyNpM6s6mhmZulleYdxM3DuINv/A3hHRMwAvggs6rf9nRExKyI6MqqfmZmVIbNRUhGxQtKUQbY/VPT0EWByVnUxM7PK1cqw2o8B3y96HsAPJQXwzYjof/fxOklXAFcAHH/88ZlW0szqm7PlVib3gCHpnRQCxhlFxWdERLekI4H7JD0dEStKHZ8Ek0UAHR0dkXmFzawuOVtu5XIdJSVpJvBPwPkRsbG3PCK6k5/rgaXA6fnU0MwahbPlVi63OwxJxwN3AZdGxK+Kyg8FRkXEtuTx2cB1OVXTzHJS7eYjZ8utXGYBQ9JtwJnAREldwBeAFoCIuAm4BpgA3CgJYE8yIuooYGlSdhDwLxHxg6zqaWa1J4vmo2PbWukuERycLTe9LEdJXTzE9o8DHy9R/jzw5gOPMLNmMVjz0XADxlXnTOsThMDZcsuVe6e3mVl/WTQfOVtu5RwwzKzmZNV85Gy5lXEuKTOrOV5sqTb5DsPMao6bj2qTA4aZ1SQ3H9UeN0mZmVkqvsMws7rm/FAjxwHDzOqW80ONLDdJmVndcn6okeWAYWZ1y/mhRpYDhpnVrYEm8jk/VDYcMMysbmU1wW/Z6m7mLFjO1Pn3MGfBcpat7q7ofI3Cnd5mVreymODnjvSBOWCYWV2r9gS/LDLlNgo3SZmZFXFH+sAcMMzMirgjfWAOGGZmRZwpd2DuwzAzK+JMuQNzwDAz68eZcktzk5SZmaXigGFmZqk4YJiZWSoOGGZmlooDhpmZpeKAYWZmqWQWMCQtlrRe0hMDbL9E0mOSHpf0kKQ3F207V9Izkn4taX5WdTQzGymNkAE3yzuMm4FzB9n+H8A7ImIG8EVgEYCk0cDXgfcAJwEXSzopw3qamWWqNwNu95adBPsz4NZb0MgsYETECmDTINsfiojNydNHgMnJ49OBX0fE8xGxG7gdOD+repqZZa1RlpKtlT6MjwHfTx63Ay8WbetKykqSdIWkTkmdGzZsyLCKZmbD0ygZcHMPGJLeSSFgfG44x0fEoojoiIiOSZMmVbdyZmZV0CgZcHMNGJJmAv8EnB8RG5PibuC4ot0mJ2VmZnWpUTLg5pZ8UNLxwF3ApRHxq6JNjwK/K2kqhUDxYeAjOVTRzKwqys2Au2x1d01my80sYEi6DTgTmCipC/gC0AIQETcB1wATgBslAexJmpb2SPoz4F5gNLA4Ip7Mqp5mZiMhbQbcWl5TXBGRawWqqaOjIzo7O/OuhpnZsM1ZsJzuEp3h7W2t/HT+u6r+epJWRkRHmn1z7/Q2M7P9anlElRdQMjOrQLX7G45tay15h1ELI6p8h2FmNkxZzOCu5RFVDhhmZsOUxQzuubPbuf6CGbS3tSIKfRfXXzAj9w5vcJOUmdmwZdXfUKtrivsOw8xsmBplBndaDhhmZsNUy/0NWXCTlJnZMJU7g7veOWCYmVWgVvsbsuAmKTMzS8UBw8zMUnHAMDOzVBwwzMwsFQcMMzNLxQHDzMxSccAwM7NUPA/DzKxOjfRSrg4YZmZ1KI+lXN0kZWZWh7JIrT4UBwwzszqUx1KuDhhmZnUoj9TqDhhmZnUoj9Tq7vQ2M6tDeaRWd8AwM6tTI51a3U1SZmaWSmYBQ9JiSeslPTHA9hMlPSzpNUl/0W/bWkmPS1ojqTOrOpqZWXpZ3mHcDJw7yPZNwCeBvxtg+zsjYlZEdFS7YmZmVr7MAkZErKAQFAbavj4iHgV6sqqDmZlVT632YQTwQ0krJV2Rd2XMzKx2R0mdERHdko4E7pP0dHLHcoAkoPQGle2ShjsvfiLwyjCPrUWNdj3QeNfUaNcDjXdNjXY9cOA1vTHtgTUZMCKiO/m5XtJS4HSgZMCIiEXAokpfU1JnI/WXNNr1QONdU6NdDzTeNTXa9UBl11RzTVKSDpU0rvcxcDZQcqSVmZmNnMzuMCTdBpwJTJTUBXwBaAGIiJskHQ10AocD+yR9GjiJwu3SUkm99fuXiPhBVvU0M7N0MgsYEXHxENt/A0wuselV4M2ZVGpwFTdr1ZhGux5ovGtqtOuBxrumRrseqOCaFBHVrIiZmTWomuvDMDOz2uSAYWZmqTR9wJB0rqRnJP1a0vy861MNjZCLq1QuMklvkHSfpGeTn+PzrGM5BrieayV1J+/TGknn5VnHckg6TtIDkn4p6UlJn0rK6/k9Guia6vJ9kjRG0s8l/SK5nv+VlE+V9LPkM+87kg5Ofc5m7sOQNBr4FfBuoAt4FLg4In6Za8UqJGkt0BERdTvhSNLbge3ALRFxSlL2t8CmiFiQBPfxEfG5POuZ1gDXcy2wPSIGyqdWsyQdAxwTEauSYfArgbnAH1O/79FA13QRdfg+qTDU9NCI2C6pBfgJ8CngfwB3RcTtkm4CfhER30hzzma/wzgd+HVEPB8Ru4HbgfNzrpMxYC6y84FvJ4+/TeGPuS4MlVut3kTEyxGxKnm8DXgKaKe+36OBrqkuRcH25GlL8i+AdwF3JOVlvUfNHjDagReLnndRx78gRRo1F9dREfFy8vg3wFF5VqZK/kzSY0mTVd003xSTNAWYDfyMBnmP+l0T1On7JGm0pDXdL/oBAAADS0lEQVTAeuA+4DlgS0TsSXYp6zOv2QNGozojIn4feA9wZdIc0lCi0JZa7+2p3wBOAGYBLwP/O9/qlE/SYcCdwKcj4tXibfX6HpW4prp9nyJib0TMojDn7XTgxErO1+wBoxs4ruj55KSsrhXn4gJ6c3E1gnVJO3Nve/P6nOtTkYhYl/xB7wP+kTp7n5J28TuBWyPirqS4rt+jUtdU7+8TQERsAR4A3gq0SeqdtF3WZ16zB4xHgd9NRg0cDHwYuDvnOlWkwXNx3Q1cljy+DPh/OdalYr0frIkPUEfvU9Kh+i3gqYj4P0Wb6vY9Guia6vV9kjRJUlvyuJXC4J6nKASODyW7lfUeNfUoKYBkiNxXgdHA4oj465yrVBFJb6JwVwH7c3HV3TUV5yID1lHIRbYMWAIcD7wAXBQRddGRPMD1nEmhmSOAtcCfFrX/1zRJZwD/DjwO7EuK/5JCm3+9vkcDXdPF1OH7JGkmhU7t0RRuDpZExHXJZ8TtwBuA1cBHI+K1VOds9oBhZmbpNHuTlJmZpeSAYWZmqThgmJlZKg4YZmaWigOGmZml4oBhVgMknSnpu3nXw2wwDhhmZpaKA4ZZGSR9NFljYI2kbybJ3bZL+kqy5sCPJE1K9p0l6ZEkad3S3qR1kn5H0v3JOgWrJJ2QnP4wSXdIelrSrcnMY7Oa4YBhlpKk6cA8YE6S0G0vcAlwKNAZEScDP6YwixvgFuBzETGTwuzh3vJbga9HxJuBP6CQ0A4K2VE/DZwEvAmYk/lFmZXhoKF3MbPEWcCpwKPJl/9WCsn19gHfSfb5v8Bdko4A2iLix0n5t4F/TfJ8tUfEUoCI2AWQnO/nEdGVPF8DTKGw6I1ZTXDAMEtPwLcj4uo+hdJf9dtvuPl2ivP57MV/n1Zj3CRllt6PgA9JOhJeX7/6jRT+jnqzf34E+ElEbAU2S3pbUn4p8ONkJbcuSXOTcxwiaeyIXoXZMPkbjFlKEfFLSZ+nsJrhKKAHuBL4LXB6sm09hX4OKKSOvikJCM8Df5KUXwp8U9J1yTkuHMHLMBs2Z6s1q5Ck7RFxWN71MMuam6TMzCwV32GYmVkqvsMwM7NUHDDMzCwVBwwzM0vFAcPMzFJxwDAzs1T+P54CqYX6kN2eAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fe59916fb00>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# list all data in history\n",
    "print(history.history.keys())\n",
    "\n",
    "# summarize history for accuracy\n",
    "plt.plot(history.history['acc'])\n",
    "plt.plot(history.history['val_acc'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()\n",
    "\n",
    "# summarize history for loss\n",
    "plt.plot(history.history['loss'],'o')\n",
    "plt.plot(history.history['val_loss'],'o')\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'val'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}