fashion-mnist.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# _________________________________________________________________\n",
    "# Layer (type)                 Output Shape              Param #   \n",
    "# =================================================================\n",
    "# input_1 (InputLayer)         (None, 1, 28, 28)         0         \n",
    "# _________________________________________________________________\n",
    "# batch_normalization_1 (Batch (None, 1, 28, 28)         112       \n",
    "# _________________________________________________________________\n",
    "# conv2d_1 (Conv2D)            (None, 64, 24, 24)        1664      \n",
    "# _________________________________________________________________\n",
    "# max_pooling2d_1 (MaxPooling2 (None, 64, 12, 12)        0         \n",
    "# _________________________________________________________________\n",
    "# conv2d_2 (Conv2D)            (None, 128, 8, 8)         204928    \n",
    "# _________________________________________________________________\n",
    "# max_pooling2d_2 (MaxPooling2 (None, 128, 4, 4)         0         \n",
    "# _________________________________________________________________\n",
    "# flatten_1 (Flatten)          (None, 2048)              0         \n",
    "# _________________________________________________________________\n",
    "# dense_1 (Dense)              (None, 256)               524544    \n",
    "# _________________________________________________________________\n",
    "# dropout_1 (Dropout)          (None, 256)               0         \n",
    "# _________________________________________________________________\n",
    "# dense_2 (Dense)              (None, 128)               32896     \n",
    "# _________________________________________________________________\n",
    "# dropout_2 (Dropout)          (None, 128)               0         \n",
    "# _________________________________________________________________\n",
    "# dense_3 (Dense)              (None, 10)                1290      \n",
    "# ================================================================="
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "2a46da40-444d-47a4-9829-94d4b05236dc",
    "_uuid": "ecba9230f6fc853da30cbdbb6b9c456d5a3a4930"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "# import essentials\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.image as mpim\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "from keras import backend as K\n",
    "K.set_image_dim_ordering('th')\n",
    "from keras.utils import np_utils\n",
    "from sklearn.model_selection import train_test_split\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Flatten, Reshape, InputLayer\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.layers.convolutional import Conv2D, MaxPooling2D"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "_cell_guid": "b574bf1b-b3a3-4e7d-9b59-e9aea6c99862",
    "_uuid": "1ca0e7d5369c09fc76ee761cc991ea1872886330",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# load data locally\n",
    "# you may want to check out loading data directly from keras\n",
    "# https://keras.io/datasets/#fashion-mnist-database-of-fashion-articles\n",
    "train_data = pd.read_csv('./data/train.csv')\n",
    "test_data = pd.read_csv('./data/test.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "_cell_guid": "e75dd0fd-2441-44b8-b628-11d61eefa3d3",
    "_uuid": "9f6d9e8e02ea6b0317bd01eb6e2b96daf5b0a154",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# build a dictionary for easy access to object classes\n",
    "objects = {0: 'T-shirt/top',\n",
    "           1: 'Trouser',\n",
    "           2: 'Pullover',\n",
    "           3: 'Dress',\n",
    "           4: 'Coat',\n",
    "           5: 'Sandal',\n",
    "           6: 'Shirt',\n",
    "           7: 'Sneaker',\n",
    "           8: 'Bag',\n",
    "           9: 'Ankle boot'}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "_cell_guid": "da11c5af-3e43-4b5b-9de1-d1e625492974",
    "_uuid": "53bfdbbc4e0393a9f1148727996bf84eef54b405"
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAADuCAYAAAAp6fzCAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2a9f221400>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# let's have a quick look of those images\n",
    "f, axes = plt.subplots(2, 4)\n",
    "for row in axes:\n",
    "    for axe in row:\n",
    "        index = np.random.randint(10000)\n",
    "        img = np.array(train_data.iloc[index, 1:]).reshape((28, 28))\n",
    "        obj = train_data.iloc[index, 0]\n",
    "        axe.imshow(img, cmap='gray')\n",
    "        axe.set_title(objects[obj])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "_cell_guid": "7abca928-b2ef-4998-8180-e0488af1d2b3",
    "_uuid": "01411f30fdc7e59c88e8af2870fe1d79047222d5",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# separate the training data into images and labels\n",
    "images = train_data.iloc[:, 1:]\n",
    "labels = train_data.iloc[:, 0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "_cell_guid": "9616dd5d-4ce9-46f2-9eeb-8d736ac05a2b",
    "_uuid": "29a563f4b1c470d802bedef9fe28d2834a38523b",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# reshape the images to fit CNNs\n",
    "labels = labels.as_matrix()\n",
    "images = images.as_matrix().reshape(images.shape[0], 1, 28, 28)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "_cell_guid": "d484a082-3b63-46a7-8739-420de46b8880",
    "_uuid": "561cd36d95a751f871ce5b6db75687a63d7c7690",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# one-hot encoding for labels\n",
    "labels = np_utils.to_categorical(labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "_cell_guid": "b6ab6d0a-f5f7-4a90-a8ab-c163bff3d1e6",
    "_uuid": "ec80d3320ba156b627be948a65570b168c01a119",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# divide data into training and validation set\n",
    "train_features, vali_features, train_labels, vali_labels = train_test_split(images, labels, \n",
    "                                                                           test_size=0.15, random_state=200)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "_cell_guid": "05913b8b-d08a-499f-b20d-7f982c6fbbee",
    "_uuid": "12545f1ec9921f7c404bec78321e444be28da958"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train_features shape:  (51000, 1, 28, 28)\n",
      "vali_features shape:  (9000, 1, 28, 28)\n",
      "train_labels shape:  (51000, 10)\n",
      "vali_labels shape:  (9000, 10)\n"
     ]
    }
   ],
   "source": [
    "# check data shapes\n",
    "print('train_features shape: ', train_features.shape)\n",
    "print('vali_features shape: ', vali_features.shape)\n",
    "print('train_labels shape: ', train_labels.shape)\n",
    "print('vali_labels shape: ', vali_labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "_cell_guid": "b86402fd-0beb-437f-ba05-c7a9bf890023",
    "_uuid": "b866ec6bf3bc0464d0a882e0c3f14b4bdcafe536",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# hyperparameters\n",
    "epochs = 30\n",
    "batch_size = 256\n",
    "\n",
    "# build model\n",
    "n_classes = 10\n",
    "model = Sequential()\n",
    "\n",
    "\n",
    "# input layer\n",
    "model.add(InputLayer(input_shape=(1, 28, 28)))\n",
    "\n",
    "# normalization\n",
    "model.add(BatchNormalization())\n",
    "\n",
    "model.add(Conv2D(64, (5, 5), activation='relu',\n",
    "          bias_initializer='RandomNormal', kernel_initializer='random_uniform'))\n",
    "\n",
    "\n",
    "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "model.add(Conv2D(512, (5, 5), activation='relu'))\n",
    "\n",
    "model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "model.add(Flatten())\n",
    "\n",
    "model.add(Dense(128, activation='relu'))\n",
    "\n",
    "model.add(Dropout(0.35))\n",
    "\n",
    "model.add(Dense(64, activation='relu'))\n",
    "\n",
    "model.add(Dropout(0.35))\n",
    "# model.add(Dense(64, activation='relu'))\n",
    "# model.add(Dropout(0.5))\n",
    "model.add(Dense(n_classes, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "_cell_guid": "dca9df43-814c-4463-9768-5427d5b312eb",
    "_uuid": "7c81ffd573b3f119dc631c597912e5dca5811e19",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from keras.optimizers import Adam\n",
    "opt = Adam(decay=0.001)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "_cell_guid": "e79248a7-8dee-40fe-91e5-3389c634c39c",
    "_uuid": "66aac56507d15659b3c23b9282acd8e2b6b0f4ba",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# compile the model\n",
    "model.compile(optimizer=opt,\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_3 (InputLayer)         (None, 1, 28, 28)         0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 1, 28, 28)         112       \n",
      "_________________________________________________________________\n",
      "conv2d_5 (Conv2D)            (None, 64, 24, 24)        1664      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_5 (MaxPooling2 (None, 64, 12, 12)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_6 (Conv2D)            (None, 512, 8, 8)         819712    \n",
      "_________________________________________________________________\n",
      "max_pooling2d_6 (MaxPooling2 (None, 512, 4, 4)         0         \n",
      "_________________________________________________________________\n",
      "flatten_3 (Flatten)          (None, 8192)              0         \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 128)               1048704   \n",
      "_________________________________________________________________\n",
      "dropout_5 (Dropout)          (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_8 (Dense)              (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "dropout_6 (Dropout)          (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_9 (Dense)              (None, 10)                650       \n",
      "=================================================================\n",
      "Total params: 1,879,098\n",
      "Trainable params: 1,879,042\n",
      "Non-trainable params: 56\n",
      "_________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "print(model.summary())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "_cell_guid": "41951e18-782f-4c61-bffe-17227b0900de",
    "_uuid": "4c355690a9be1b8f6d7a64d9734c8d2e6d8030c0",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 51000 samples, validate on 9000 samples\n",
      "Epoch 1/30\n",
      "51000/51000 [==============================] - 25s - loss: 0.7554 - acc: 0.7336 - val_loss: 0.3804 - val_acc: 0.8639\n",
      "Epoch 2/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.4353 - acc: 0.8511 - val_loss: 0.3099 - val_acc: 0.8904\n",
      "Epoch 3/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.3608 - acc: 0.8767 - val_loss: 0.2810 - val_acc: 0.8993\n",
      "Epoch 4/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.3156 - acc: 0.8936 - val_loss: 0.2528 - val_acc: 0.9077\n",
      "Epoch 5/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.2816 - acc: 0.9036 - val_loss: 0.2407 - val_acc: 0.9133\n",
      "Epoch 6/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.2525 - acc: 0.9128 - val_loss: 0.2289 - val_acc: 0.9209\n",
      "Epoch 7/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.2293 - acc: 0.9200 - val_loss: 0.2241 - val_acc: 0.9220\n",
      "Epoch 8/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.2090 - acc: 0.9272 - val_loss: 0.2223 - val_acc: 0.9222\n",
      "Epoch 9/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.1934 - acc: 0.9317 - val_loss: 0.2134 - val_acc: 0.9246\n",
      "Epoch 10/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.1780 - acc: 0.9377 - val_loss: 0.2217 - val_acc: 0.9244\n",
      "Epoch 11/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.1656 - acc: 0.9408 - val_loss: 0.2176 - val_acc: 0.9270\n",
      "Epoch 12/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.1531 - acc: 0.9461 - val_loss: 0.2169 - val_acc: 0.9292\n",
      "Epoch 13/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.1402 - acc: 0.9509 - val_loss: 0.2127 - val_acc: 0.9297\n",
      "Epoch 14/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.1306 - acc: 0.9542 - val_loss: 0.2214 - val_acc: 0.9268\n",
      "Epoch 15/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.1179 - acc: 0.9586 - val_loss: 0.2284 - val_acc: 0.9264\n",
      "Epoch 16/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.1103 - acc: 0.9620 - val_loss: 0.2292 - val_acc: 0.9288\n",
      "Epoch 17/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0999 - acc: 0.9657 - val_loss: 0.2278 - val_acc: 0.9307\n",
      "Epoch 18/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0912 - acc: 0.9678 - val_loss: 0.2432 - val_acc: 0.9268\n",
      "Epoch 19/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0854 - acc: 0.9699 - val_loss: 0.2447 - val_acc: 0.9301\n",
      "Epoch 20/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0777 - acc: 0.9736 - val_loss: 0.2572 - val_acc: 0.9310\n",
      "Epoch 21/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0715 - acc: 0.9749 - val_loss: 0.2666 - val_acc: 0.9293\n",
      "Epoch 22/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0669 - acc: 0.9759 - val_loss: 0.2634 - val_acc: 0.9302\n",
      "Epoch 23/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0610 - acc: 0.9785 - val_loss: 0.2649 - val_acc: 0.9319\n",
      "Epoch 24/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0582 - acc: 0.9798 - val_loss: 0.2684 - val_acc: 0.9308\n",
      "Epoch 25/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0543 - acc: 0.9801 - val_loss: 0.2772 - val_acc: 0.9286\n",
      "Epoch 26/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0488 - acc: 0.9829 - val_loss: 0.2917 - val_acc: 0.9296\n",
      "Epoch 27/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0465 - acc: 0.9842 - val_loss: 0.2815 - val_acc: 0.9293\n",
      "Epoch 28/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0416 - acc: 0.9850 - val_loss: 0.2942 - val_acc: 0.9293\n",
      "Epoch 29/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0392 - acc: 0.9860 - val_loss: 0.3126 - val_acc: 0.9314\n",
      "Epoch 30/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0357 - acc: 0.9876 - val_loss: 0.3048 - val_acc: 0.9316\n"
     ]
    }
   ],
   "source": [
    "# training\n",
    "import time\n",
    "start_time = time.time()\n",
    "\n",
    "training = model.fit(train_features, train_labels,\n",
    "                     validation_data=(vali_features, vali_labels),\n",
    "                     epochs=epochs,\n",
    "                     batch_size=batch_size, \n",
    "                     verbose=1)\n",
    "training_time = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training 30 epochs in 12:11.4\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEWCAYAAACJ0YulAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nbp2pJlX2EJIAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2a5c05a4a8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYsAAAEWCAYAAACXGLsWAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3Xl4VdW9//H3NyEhAUICJMwzggwiIClOOFXrgLdSrW3F\ntk7t5dafQ9tbO147aOut1WodOzjW9qpcW69D73Ws4lSrzCCDzAgkARJCApmn7++PvZMcQsgAOTkZ\nPq/n2c85Z5+991k7gfPJXmuvtczdERERaUpcrAsgIiIdn8JCRESapbAQEZFmKSxERKRZCgsREWmW\nwkJERJqlsBARkWYpLKTbM7O3zGyfmfWMdVlEOiqFhXRrZjYaOA1w4KJ2/Nwe7fVZIm1BYSHd3RXA\nB8AfgStrV5pZspndZWafmFmhmb1nZsnhe7PN7H0zKzCzHWZ2Vbj+LTP7esQxrjKz9yJeu5ldZ2Yb\ngY3hunvDY+w3s6VmdlrE9vFm9iMz22xmB8L3R5jZg2Z2V+RJmNmLZvbtaPyAREBhIXIF8GS4nGdm\ng8L1vwZmAqcA/YHvATVmNgp4GbgfyACmAyta8XmfA04EJoevF4fH6A88BfzFzJLC9/4dmAfMAfoC\n1wAlwBPAPDOLAzCzdOCccH+RqFBYSLdlZrOBUcAz7r4U2AxcHn4JXwN8092z3L3a3d9393LgcuDv\n7v60u1e6+153b01Y/NLd8929FMDd/ys8RpW73wX0BI4Nt/06cLO7r/fAynDbRUAhcHa43WXAW+6+\n+yh/JCKHpbCQ7uxK4DV3zwtfPxWuSweSCMKjoRGHWd9SOyJfmNlNZrYurOoqAFLDz2/us54AvhI+\n/wrw56Mok0iz1Mgm3VLY/vBFIN7MdoWrewJpwBCgDBgHrGyw6w5g1mEOWwz0ing9uJFt6oZ5Dtsn\nvkdwhbDG3WvMbB9gEZ81DljdyHH+C1htZtOAScDzhymTSJvQlYV0V58DqgnaDqaHyyTgXYJ2jMeA\nu81saNjQfHJ4a+2TwDlm9kUz62FmA8xsenjMFcAlZtbLzI4BvtZMGVKAKiAX6GFmPyFom6j1CPBz\nMxtvgePNbACAu+8kaO/4M/BsbbWWSLQoLKS7uhJ43N23u/uu2gV4APgy8APgI4Iv5HzgV0Ccu28n\naHD+Trh+BTAtPOZvgApgN0E10ZPNlOFV4BVgA/AJwdVMZDXV3cAzwGvAfuBRIDni/SeAqagKStqB\nafIjkc7JzE4nqI4a5fqPLFGmKwuRTsjMEoBvAo8oKKQ9RC0szOwxM9tjZo01zhHWwd5nZpvMbJWZ\nnRDx3pVmtjFcrmxsf5HuyswmAQUEDfH3xLg40k1ErRoqvEQuAv7k7sc18v4c4AaC+t8TgXvd/UQz\n6w8sATIJ7hxZCsx0931RKaiIiDQralcW7v4OQQPg4cwlCBJ39w+ANDMbApwHvB52XNoHvA6cH61y\niohI82LZz2IYB9/5sTNcd7j1TUpPT/fRo0e3ZflERLq8pUuX5rl7RnPbdepOeWY2H5gPMHLkSJYs\nWRLjEomIdC5m9klLtovl3VBZBMMZ1Boerjvc+kO4+0PununumRkZzQajiIgcoViGxYvAFeFdUScB\nhe6eQ9BR6Vwz62dm/YBzw3UiIhIjUauGMrOngTOBdDPbCfwUSABw998DLxHcCbWJYNjlq8P38s3s\n5wQ9ZwFudfemGspFRCTKohYW7j6vmfcduO4w7z1GMDaPiIh0AOrBLSIizVJYiIhIsxQWIiLSrE7d\nz0JEpLupqXHyisrJKiglq6CU7IJS+vRM4PITR0b1cxUWIiIdRFllNfnFFeQXV7C3uIKcMAyyCsrI\nKighu6CMnMJSKqsPHtPvhJFpCgsRkc6kqrqGovIqDpRVsb+skgNlVeESPC8srSS/uIJ9JRV1j/uK\ng3WlldWHHC/OYHDfJIamJTN9RBpzpg5hWFrwunZJTU6I+nkpLEREmlBd4+wtLifvQAW5ReXkHqhf\n8orqH2uDoaTi0C/8hlKSetC/dyL9eiUyMCWJYwf1pX/vBPr1TqR/r8TgsXciQ1KTGNQ3iYT42Dcv\nKyxERAB3J6uglBU7ClixvYCVOwvYmldCfnE5NY3M5NA7MZ6MlJ5kpPRkXEYfUpMTSEnqQUpS7WPw\nvG/42Cdc1zcpgcQesf/yby2FhYh0S/vLKlm1o5AVO/YFAbGjkLyicgASe8Rx3NC+nDNpIAPDQKhb\n+iSRnpJIr8Tu9fXZvc5WRLqEmhpnX0kFew6UsyesEjpQVklVtVNZU0NVtVNV41RV11BV41RW11Bd\n41RWOyUVVazJ3s/m3CJq534bm9Gb0yekM2NEGtNGpDFxcN9O+dd/NCksRKTDKa2oZt2u/Xycc4Bd\nhaXkFpWzZ3953WNeUTlVjdUNNdAjzugRbyTExdEj3ugRH0fPHnEcOyiFudOGMn1kGscPT2uXBuLO\nTmEhIjFVVF7F2uz9rM4qZHV2IauzCtm0p6iuncAMBvTuWVcddOygFDJSal8nMbBvTzL69CQ1OSEI\nhvg4esQZ8XGGmcX25LoQhYWIRFVFVQ0FpRUUlFSyr7iCgtJKtu8t4aMwHLbmFddVB2Wk9GTqsFTO\nnzKYKcNSmTK0L4P7JtGjA9wN1N0pLESkVUorqskrCtoKIm8dzSsqZ19JJQUlQTAUhM+LD3Mr6dDU\nJKYMS+Vz04dx3LC+HDc0lYF9k9r5bKSlFBYicojqGmfVzgLe3pDL+l0HDupTcLgv/7ReCfTvlUha\nrwQG9U3i2MEp9OuVSFpyAmm9g8d+4ftDUpMY0KdnO5+VHA2FhYgAsHt/GW9vyOXtDbn8Y1MeBSWV\nmMGY9N4MTOnJ1OFppPdJJCOlJ+l9gnaC2ucD+iR2iI5jEj0KC5FuqryqmiXb9vH2hlze2ZDLx7sO\nADAwpSfnTBrEGRMymH1MOv16J8a4pNIRKCxEuoF9xRVsySti855iNucW8fGuAyzamk9pZTUJ8can\nRvfnhxdM5PQJGUwcnKK7iOQQCguRLqK6xtmRX3JQKARLMfnFFXXbJfaIY2x6b76QOZwzJmRw0tgB\n9O6prwJpmv6FiHQi7s6+kkq25BaxJbeYzXlFbM0tZkteMdv3llBRXVO3bXqfRMZm9OG8KYMYl9Gn\nbhnWL5n4OF05SOsoLEQ6qJoaZ3V2If/YtJdNe4rYkhcERGFpZd02CfHGqAG9GZvem3MmDWJsRu8w\nFHqT1kttDdJ2FBYiHcieA2W8uyGPdzbm8u7GvLrqo0F9ezI2vQ//cvwQxqQHgTA2ozfD0pLVYU3a\nhcJCJIYqqmpY+kn9HUlrc/YDQRXSGRMygjuSxqeTrj4JEmMKC5F2VFPjbNxTxD835/Hepr38c3Me\nxRXV9IgzThjVj++edyxnTMhg8pC+xKldQToQhYVIFLkH4fDBlr3hkl9XtTS8XzKfmzGM0ydkcMq4\nAaQkaeRT6bgUFiJtyN3ZnFvEP7fk88HmICD2huEwNDWJM48NblU9eewARvTvFePSirScwkLkKFVW\n1/DBlr28vHoXr6/dTe6BYLa1IalJdf0YTho7gBH9k9XZTTothYXIESivqua9jXm8vHoXf1+3m4KS\nSnolxnPWsQM5fUI6J40dwMj+vRQO0mUoLERaqLSimrfW7+Hl1bt48+M9FJVXkZLUg3MmDeL84wZz\nxoQMkhLiY11MkahQWIgchruzPb+ED7fk8+bHe3hrwx7KKmvo1yuBC6cO4fypgzl1XLrmapZuIaph\nYWbnA/cC8cAj7n57g/dHAY8BGUA+8BV33xm+Vw18FG663d0vimZZRWpqnE25RXy4NZ9FW/NZtHUv\nu/cH7Q8ZKT35wswRXHDcYGaN6a+OcNLtRC0szCweeBD4DLATWGxmL7r72ojNfg38yd2fMLNPA78E\nvhq+V+ru06NVPpHqGmddzv4wHPayaGs++0qCoTQG9e3JiWMGMGtMf04c059xGX3U70G6tWheWcwC\nNrn7FgAzWwDMBSLDYjLw7+HzhcDzUSyPCDU1zuJt+Ty3PIuXPsphf1kVACP79+KcSYPCcNCdSyIN\nRTMshgE7Il7vBE5ssM1K4BKCqqqLgRQzG+Due4EkM1sCVAG3u7uCRI7Ypj1FPLd8J88vzyaroJRe\nifGcP2UwZxybwawx/RmSmhzrIop0aLFu4L4JeMDMrgLeAbKA2gl+R7l7lpmNBd40s4/cfXPkzmY2\nH5gPMHLkyPYrtXQKeUXlvLgim+eWZ/FRViFxBqeNz+C75x3LuVMG0Ssx1v/8RTqPaP5vyQJGRLwe\nHq6r4+7ZBFcWmFkf4PPuXhC+lxU+bjGzt4AZwOYG+z8EPASQmZnpUTkL6VRKK6p5be0unluexbsb\n86iucaYM7cvNF07ioulDGZiSFOsiinRK0QyLxcB4MxtDEBKXAZdHbmBm6UC+u9cAPyS4Mwoz6weU\nuHt5uM2pwB1RLKt0cquzCvnvxTt4fkUWB8qqGJqaxPzTx3LxjGFMGJQS6+KJdHpRCwt3rzKz64FX\nCW6dfczd15jZrcASd38ROBP4pZk5QTXUdeHuk4A/mFkNEEfQZrH2kA+Rbu1AWSUvrsxmwaIdfJRV\nSGKPOOYcN5gvZo7gpLEDdPeSSBsy965Re5OZmelLliyJdTEkytydZdv3sWDRDv53VQ6lldVMHJzC\nZZ8awedmDNPscCKtZGZL3T2zue3Uwiedwr7iCp5dtpP/XryDjXuK6JUYz9zpQ7ls1kimDU/Vba4i\nUaawkA7H3ckuLGPJtnyWfbKPpdv3sS7nANU1zvQRadx+yVT+ZdpQ+vTUP1+R9qL/bRJzldU1rMvZ\nz5JtQTAs3baPXfvLAOiVGM/0EWn8vzPHMWfqECYN6Rvj0op0TwoLiYmi8iqeWbyD19buYuWOQkor\ng+41w9KSmTWmPzNH9WPmqH5MHJyicZhEOgCFhbSrrIJS/viPrSxYtIMD5VVMHtKXy2aNqAsH9aQW\n6ZgUFtIuVuwo4JF3t/Dy6l0AXDh1CF+bPYZpI9JiXDIRaQmFhURNdY3z+tpdPPLuVpZ8so+UpB58\nffYYrjhlNMPSdAUh0pkoLKTN1bZHPP7+VnbklzKifzI//exkvpA5QncwSctVVsKePZCSEiy6PTqm\n9D9X2kR5VTXvbMjjpY9yeH3tborKq8gc1Y//mDOJz0weTLx6Ux+ZkhLIyoIxY6BHFP+7VldDQQHs\n21f/WLvEx8OoUTB6NIwcCT17ts1nlpfDjh3wySewbVuw1D7/5BPYuRNqaoJtExMhI6PpZcQImDgR\n+vRpm/K1p4oK2LwZ1q2Djz8Ols2bg/U1NY0v1dX1z6dNg+eei2oRFRZyxMqrqnk3IiAOlFeRmpzA\nBccN5vITRzJjZL/2L1RNDRQWQmkplJU1/VhZGXy5nHACJMewWmz/fti0qX7ZvLn+eXZ2sE1aGpx7\nLsyZA+efD4MGtf5zSkvhn/+Et9+GDz4I/mqvDYT9+1t+nCFD6sNj9Oj658OGBT/byKBpGDy163bt\nCs4tcgSJuDgYPjw41hln1B+zqAhyc4Nlz57gcfPm4PHAgUPLN3o0TJ588DJpEvRt5rZr9+B4e/fW\nL2VlQUgnJASPkUvDdS258nGHvLz6QIgMhurq+u2GD4djjoHU1CCs4+IaX2rfGzeu+c8+ShruQ1ql\nvKqa9zbm8X8f5fD6mvqAOHfyIC48fgintNec1NXVsHUrrF1bv6xZE/zHKylp3bF69IDjj4cTT6xf\nJkwI/hO2FXfIyYHVq+uXjz8OAiE39+BtBw8Ovihql8GD4f334aWXgi9ZgMzMIDjmzAmex8cf+pnF\nxUE4vPVWEBCLFgV/qcbFBec7YkQQQv36Nb1UVjb+1/+2bcGVQWVl0+eemHjoMQcObDxsEhJa93Mt\nKwu+fHNzg/LU/jtYuzb4+ZaX1287fDhMmRL8TMvKDg6FvXshP7/5c2lLiYkwfnzwB8vEiUGgTZwY\n/NtLab/BL1s63IfCQlpk6Sf7eOrD7by2dhcHyqrom9SD86YMZs7xQzg1mgHhHnwJrFxZ/yVQ+0VQ\nVla/3fDh9X9FjhoVXCkkJ0NSUv1j5PPk5OAvwY8+gg8/DJbFi+v/Uk1NhU99qj48jj022C8xsX7p\n2bPxvyj37QvKunp1cPzacMjPr99m0KDgy2H8+IODYezYw1ejuMOKFUFovPxyEAQ1NTBgQHC1MWcO\n9O8P77wTBMTixVBVFQTJCSfAmWcGf7HPnh2cX1uorg5CcNu2oLqsV69DgyFWV22Rf1BE/tvZtAl6\n9w5+bs0tycnBz7DhUll56OuWSksLQmH06OhWLbaQwkLaxIbdB7jjlfX8fd1uUsKAuHDqEE49JgoB\nUVkZ1NkuXx58KdY+FhbWbzNqVH0oTJnS8iqGlqiuhvXr68Pjww+DL/vI6oHG1AZHYmIQHHl59e+l\npMBxx8HUqcFj7ZKRcfTlzc+H116rD4/az+3RI7jaqA2HU09t179UpXNRWMhRySoo5Tevb+B/lu2k\nd2IP/u2MsVwze0zTs8vt21dfvVJb3RFZr9rYkpdXHwqrVwf7QfAX3fHHw4wZMH16sEyZ0v6NlyUl\nsHRpUPVSUVG/lJcf/Fj7vKoquDqoDYcRI9rnLp6aGliyJLgyOumk4C9nkRZQWMgR2VdcwYMLN/Gn\nDz4B4MqTR/H/zjyGfr0jhv4uKQku52urV2qrWmobY1srPb0+FGofJ0xovB5eRNqUhiiXQ1VWBn/J\n195RkpsbNOyVl1NRVs6yzbks25xLWlUVv8/oxaeGp5DyGvBSWC+7Z08QCps319/FkpQUVAWdc059\ndcukSUHddWO3+DVcUlJg6FDdQy/SwSksuhp3eOGFoB47MhRyc4NbFg8jETgpXDw+HmvsNsG0tOCv\n/q9+tb7ufdw4XQGIdAMKi65k8WL4znfg3XeDu2KGDg0aUmfMCB4HDqzrwFQzIJ239sF9K/PZuL+a\n6WMy+Pc5k5g5NgNry1tGRaRLUFh0Bdu3w49+BE8+GQTC738PX/tao7fluTtvb8jljlfWszZnPxMH\np3P/vGM569iBmm1ORA5LYdGZHTgAt98Od98dVD/98Ifwgx8c9jbSpZ/s445XPubDrfmM6J/MPV+a\nzkXThhKnoThEpBkKi86oqgoeewx+/OOgXeLLX4b//M9g3J5GbNh9gDtfXc/ra3eT3qcnt86dwmWf\nGtk+Pa1FpEtQWHQ2r74atEusWRP0xP3b32DWrEY33ZFfwm/+voHnlmfRJ7EH3z3vWK4+dXTTfSVE\nRBqhb43OoKYG/v53uOuuoMfuuHHw7LNw8cWN3nKaX1zBfW9s5MkPPyHOjPmnjeUbZ4w7uK+EiEgr\nKCw6ssJCeOIJePBB2LAhuJPprrvg+uuDoSUa8crqHP7judUUlFbyxczh3Hj2eE1VKiJHTWHREa1e\nHQTEn/8cjBx60knB8y984bBzCRSWVPLTF1fz/IpsjhvWl6f+9SSOHazxgESkbSgsOorKSnj++SAk\n3n47CIXLL4frroOZM5vcdeH6Pfzg2VXsLargW+eM57qzjiEhXo3XItJ2FBaxlpcHv/td0DciOzsY\ntviOO+Caa4IhkptwoKyS2/5vHQsW72DCoD48euWnOG5YGw09LSISQWERKzU1we2v3/teMFrreecF\ngTFnTouGz3h/cx7f/csqcgpL+cYZ4/j2Z8bTs4eG3RCR6FBYxMJHH8G118I//gGnnx5UPR13XIt2\nLa2o5levfMwf39/G6AG9+Ms3TmbmqP5RLrCIdHcKi/ZUXAy33hr0uE5NhT/+Ea64osUjri79ZB83\n/WUlW/OKueqU0Xzv/GPVZ0JE2kVUW0HN7HwzW29mm8zsB428P8rM3jCzVWb2lpkNj3jvSjPbGC5X\nRrOc7eJvfwuG8r7jDrjyymBGtiuvbFFQVNc497+xkS/8/n0qqmp46usn8rOLpigoRKTdRO3bxszi\ngQeBzwA7gcVm9qK7r43Y7NfAn9z9CTP7NPBL4Ktm1h/4KZAJOLA03HdftMobNTt2wI03Bnc6TZkS\njAg7e3aLd889UM63/3sF723K46JpQ7nt4uNISWrlpPYiIkcpmlcWs4BN7r7F3SuABcDcBttMBt4M\nny+MeP884HV3zw8D4nXg/CiWte1VVgYd6CZNCobouP12WLasVUHxj015XHDvuyzels+vPj+Vey+b\nrqAQkZiIZj3GMGBHxOudwIkNtlkJXALcC1wMpJjZgMPsO6zhB5jZfGA+wMjDDKIXE0uXBkOEr1wJ\nF14IDzwQ3BLbQtU1zr1vbOT+NzcyLqMPT379RHWwE5GYinXPrZuAM8xsOXAGkAVUt3Rnd3/I3TPd\nPTMjIyNaZWy5srJgXokTTwxGg3322aCtohVBsXt/GZc//AH3vbGRz58wnBevP1VBISIxF80riyxg\nRMTr4eG6Ou6eTXBlgZn1AT7v7gVmlgWc2WDft6JY1qP34Ydw9dWwbl3wePfdwTSkrfD2hly+/d8r\nKK2o5q4vTOPzM4c3v5OISDuI5pXFYmC8mY0xs0TgMuDFyA3MLN3MasvwQ+Cx8PmrwLlm1s/M+gHn\nhus6ntJSuOkmOOUUKCqCl18OOtu1Iiiqqmv41Ssfc+Vji8jo05O/3TBbQSEiHUrUrizcvcrMrif4\nko8HHnP3NWZ2K7DE3V8kuHr4pZk58A5wXbhvvpn9nCBwAG519/xolfWIvfdeMCzHxo3wb/8W3BZ7\nmFnqDmdXYRnXP7WMJZ/sY96sEfz0s1NISlBPbBHpWMzdY12GNpGZmelLlixpnw8rLg7aJu6/H0aN\ngkcegbPPbvVhisqruPjBf5BdUMp/XjKVudMPacMXEYkqM1vq7pnNbadeXa21cGFwp9PWrXDDDcF0\npn36tPow7s5Nz6xkc24Rf/7aiZx6THoUCisi0jZifTdU5/LAA/DpTwcD/b3zDtx33xEFBcBv39rM\nK2t28aM5kxQUItLh6cqipQoK4Oab4TOfCXpj9+p1xIdauH4Pv35tPRdNG8rXZo9pw0KKiESHrixa\n6je/CaY5vfPOowqKbXnFfPPp5Uwc3Jdfff54rIWDCIqIxJLCoiXy84OwuPRSmDbtiA9TXF7Fv/15\nKXFxxkNfnUlyou56EpHOQWHREnfdFfSh+OlPj/gQ7s73/rqKjXsOcP+8GYzof+RXJyIi7U1h0Zy8\nPLj3XvjSl1o8QVFj/vDOFv7voxy+f/5EThvfAYYmERFpBYVFc+68M+ilfRRXFe9syOWOVz7mwuOH\nMP/0sW1YOBGR9qGwaMru3cHtspdfDhMnHtEhtu8t4YanlzNhUAp3XqoGbRHpnBQWTbnjDigvh5/8\n5Ih2L6moYv6fg17lf/jqTM1sJyKdlr69DicnB377W/jqV2H8+Fbv7u58/9mPWL/7AH+8ehajBvSO\nQiFFRNqHriwO5/bbg9nufvzjI9r94Xe38LeV2Xz3vGM5Y4IatEWkc1NYNGbnTvjDH4J5Kca2vkF6\nw+4D3P7yx1xw3GCuPWNcFAooItK+FBaN+eUvoaYG/uM/jmj3X7+6nt6JPfjPi6eqQVtEugSFRUPb\nt8PDDwcjy7ZiOtRay7fv47W1u/nX08fSr3di25dPRCQGFBYN3XYbmAXzVRyBO19dz4DeiVyjAQJF\npAtRWETaujWYEnX+fBgxovntG3hvYx7vb97LdWcdQ5+eutFMRLqOFoWFmY0zs57h8zPN7EYza/kk\n053FL34RzFXxwx+2eld3585XP2ZYWjJfPmlkFAonIhI7Lb2yeBaoNrNjgIeAEcBTUStVLGzaBE88\nAddeC0OHtnr3V9fsYuXOQr55znh69tBosiLStbQ0LGrcvQq4GLjf3b8LDIlesWLg5z+HxET4/vdb\nvWt1jfPr1zZwzMA+XDJD82iLSNfT0rCoNLN5wJXA/4brEqJTpBhYvx7+67/guutg8OBW7/4/y3ay\naU8RN507gR7xagYSka6npd9sVwMnA7e5+1YzGwP8OXrFame33grJyfC977V61/Kqau75+0amDU/l\nvCmtDxoRkc6gRbfsuPta4EYAM+sHpLj7r6JZsHazcSM8/XRQ/ZTR+mE5nvpwO1kFpZoiVUS6tBaF\nhZm9BVwUbr8U2GNm/3D3f49i2drHMcfACy/AySe3etfi8ioeeHMTp4wbwOzx6VEonIhIx9DSaqhU\nd98PXAL8yd1PBM6JXrHakRl89rOQ3vov+8fe28re4gq+e96xUSiYiEjH0dKw6GFmQ4AvUt/A3a3t\nK67goXe2cO7kQcwY2S/WxRERiaqWhsWtwKvAZndfbGZjgY3RK1bH9/u3N1NUUcVNuqoQkW6gpQ3c\nfwH+EvF6C/D5aBWqo9tVWMYf39/GxTOGMWFQSqyLIyISdS0d7mO4mT1nZnvC5VkzGx7twnVU976x\nkRp3vn3OhFgXRUSkXbS0Gupx4EVgaLj8LVzX7WzNK+aZJTu4fNZIRvTvFeviiIi0i5aGRYa7P+7u\nVeHyR6DZTglmdr6ZrTezTWb2g0beH2lmC81suZmtMrM54frRZlZqZivC5fetOqsouvv1DSTGx3H9\np1s/L7eISGfV0nG095rZV4Cnw9fzgL1N7WBm8cCDwGeAncBiM3sx7OBX62bgGXf/nZlNBl4CRofv\nbXb36S0sX7tYk13I31Zmc91Z48hI6Rnr4oiItJuWXllcQ3Db7C4gB7gUuKqZfWYBm9x9i7tXAAuA\nuQ22caBv+DwVyG5heWJiwaId9E6MZ/7pmldbRLqXFoWFu3/i7he5e4a7D3T3z9H83VDDgB0Rr3eG\n6yL9DPiKme0kuKq4IeK9MWH11NtmdlpjH2Bm881siZktyc3NbcmpHJXt+SWMzehDanLXGUNRRKQl\njmaI1LYY6mMe8Ed3Hw7MAf5sZnEEVy8j3X1G+DlPmVnfhju7+0PununumRlHMK5Ta+UUljIkNSnq\nnyMi0tEcTVg0N2peFsEkSbWGh+sifQ14BsDd/wkkAenuXu7ue8P1S4HNQMzvU80uKGNoWnKsiyEi\n0u6OJiy8mfcXA+PNbIyZJQKXEdx+G2k7cDaAmU0iCItcM8sIG8gJe4uPB7YcRVmP2v6ySorKqxia\npisLEelGykDFAAATyElEQVR+mrwbyswO0HgoGNDkn9juXmVm1xMMExIPPObua8zsVmCJu78IfAd4\n2My+HX7OVe7uZnY6cKuZVQI1wDfcPb+1J9eWcgrKABiSqisLEel+mgwLdz+qsSzc/SWChuvIdT+J\neL4WOLWR/Z4lmPe7w8guKAVQNZSIdEuaA7SFsgtrw0LVUCLS/SgsWiinoIz4OGNgisJCRLofhUUL\nZReUMrhvEvFxmjpVRLofhUULZauPhYh0YwqLFlIfCxHpzhQWLVBT4+wqLGOIGrdFpJtSWLTA3uIK\nKqprGKo+FiLSTSksWkB9LESku1NYtEBO2MdCDdwi0l0pLFogOxzqY5iuLESkm1JYtEB2QSlJCXGk\n9dI8FiLSPSksWiCnsIyhqcmYqUOeiHRPCosWyCooVeO2iHRrCosW0Ax5ItLdKSyaUVldw54D5QzR\nlYWIdGMKi2bsKizDHYap97aIdGMKi2bkFGqGPBERhUUz1HtbRERh0SzNkCciorBoVk5BGanJCfRK\nbHK6chGRLk1h0Yxs9bEQEVFYNCe7sIyh6mMhIt2cwqIZOYWlmvRIRLo9hUUTSiqqKCipVDWUiHR7\nCosm1A5NrhnyRKS7U1g0QX0sREQCCosmaIY8EZGAwqIJ2QVlmMFghYWIdHMKiyZkF5QyMKUnCfH6\nMYlI96ZvwSbkFJZpAEEREaIcFmZ2vpmtN7NNZvaDRt4faWYLzWy5ma0yszkR7/0w3G+9mZ0XzXIe\nTtB7W1VQIiJRCwsziwceBC4AJgPzzGxyg81uBp5x9xnAZcBvw30nh6+nAOcDvw2P127cnezCUt02\nKyJCdK8sZgGb3H2Lu1cAC4C5DbZxoG/4PBXIDp/PBRa4e7m7bwU2hcdrNwUllZRV1miGPBERohsW\nw4AdEa93husi/Qz4ipntBF4CbmjFvpjZfDNbYmZLcnNz26rcAGSFfSw0Q56ISOwbuOcBf3T34cAc\n4M9m1uIyuftD7p7p7pkZGRltWjDNkCciUi+akzRkASMiXg8P10X6GkGbBO7+TzNLAtJbuG9U1XXI\n05WFiEhUrywWA+PNbIyZJRI0WL/YYJvtwNkAZjYJSAJyw+0uM7OeZjYGGA8simJZD5FVUEpifBzp\nvXu258eKiHRIUbuycPcqM7seeBWIBx5z9zVmdiuwxN1fBL4DPGxm3yZo7L7K3R1YY2bPAGuBKuA6\nd6+OVlkbk1NQxuDUJOLirD0/VkSkQ4rqXKHu/hJBw3Xkup9EPF8LnHqYfW8Dbotm+ZqSXVCqMaFE\nREKxbuDusHIKyxim22ZFRACFRaOqa5xd+8vUuC0iElJYNGLPgTKqa1zzWIiIhBQWjdAMeSIiB1NY\nNKJ2hjxVQ4mIBBQWjajtkKdqKBGRgMKiEdkFZfTp2YO+SQmxLoqISIegsGiE+liIiBxMYdGInMIy\nVUGJiERQWDQip1Az5ImIRFJYNFBWWU1eUYWGJhcRiaCwaGBXOI+FqqFEROopLBqo7WMxVA3cIiJ1\nFBYNZOvKQkTkEAqLBnLCK4vBurIQEamjsGggu7CUAb0TSUqIj3VRREQ6DIVFA9kF6mMhItKQwqIB\n9d4WETlUVKdV7YxyCss49Zj0WBdDpFurrKxk586dlJWVxbooXUZSUhLDhw8nIeHIxrxTWETYX1ZJ\nUXmVem+LxNjOnTtJSUlh9OjRmFmsi9PpuTt79+5l586djBkz5oiOoWqoCHXzWKj3tkhMlZWVMWDA\nAAVFGzEzBgwYcFRXagqLCDkF6mMh0lEoKNrW0f48FRYRsusmPVI1lIhIJIVFhOyCUuLjjIEpCguR\n7mzv3r1Mnz6d6dOnM3jwYIYNG1b3uqKiokXHuPrqq1m/fn2T2zz44IM8+eSTbVHkqFMDd4ScgjIG\n900iPk6XvyLd2YABA1ixYgUAP/vZz+jTpw833XTTQdu4O+5OXFzjf3M//vjjzX7Oddddd/SFbScK\niwhZ6mMh0uHc8rc1rM3e36bHnDy0Lz/97JRW77dp0yYuuugiZsyYwfLly3n99de55ZZbWLZsGaWl\npXzpS1/iJz/5CQCzZ8/mgQce4LjjjiM9PZ1vfOMbvPzyy/Tq1YsXXniBgQMHcvPNN5Oens63vvUt\nZs+ezezZs3nzzTcpLCzk8ccf55RTTqG4uJgrrriCdevWMXnyZLZt28YjjzzC9OnT2/Rn0hxVQ0XI\nKSxjiBq3RaQJH3/8Md/+9rdZu3Ytw4YN4/bbb2fJkiWsXLmS119/nbVr1x6yT2FhIWeccQYrV67k\n5JNP5rHHHmv02O7OokWLuPPOO7n11lsBuP/++xk8eDBr167lxz/+McuXL4/q+R2OrixCNTXOrsIy\nLpiqKwuRjuRIrgCiady4cWRmZta9fvrpp3n00UepqqoiOzubtWvXMnny5IP2SU5O5oILLgBg5syZ\nvPvuu40e+5JLLqnbZtu2bQC89957fP/73wdg2rRpTJkSm5+HwiKUV1xORXUNQ9XHQkSa0Lt377rn\nGzdu5N5772XRokWkpaXxla98pdG+DImJiXXP4+PjqaqqavTYPXv2bHabWFE1VEh9LESktfbv309K\nSgp9+/YlJyeHV199tc0/49RTT+WZZ54B4KOPPmq0mqs9RPXKwszOB+4F4oFH3P32Bu//BjgrfNkL\nGOjuaeF71cBH4Xvb3f2iaJY1p7C297aqoUSkZU444QQmT57MxIkTGTVqFKeeemqbf8YNN9zAFVdc\nweTJk+uW1NTUNv+c5pi7R+fAZvHABuAzwE5gMTDP3RuNRTO7AZjh7teEr4vcvU9LPy8zM9OXLFly\nxOV99L2t/Px/17Lsx5+hf+/E5ncQkahZt24dkyZNinUxOoSqqiqqqqpISkpi48aNnHvuuWzcuJEe\nPVr/t35jP1czW+rumYfZpU40ryxmAZvcfUtYoAXAXOBw11DzgJ9GsTxNyikoJSkhjn69jmxERhGR\naCgqKuLss8+mqqoKd+cPf/jDEQXF0YrmJw4DdkS83gmc2NiGZjYKGAO8GbE6ycyWAFXA7e7+fCP7\nzQfmA4wcOfKoCptdWMrQ1GSNRyMiHUpaWhpLly6NdTE6TAP3ZcBf3b06Yt2o8NLocuAeMxvXcCd3\nf8jdM909MyMj46gKkF1QxhCNCSUi0qhohkUWMCLi9fBwXWMuA56OXOHuWeHjFuAtYEbbF7FeTnhl\nISIih4pmWCwGxpvZGDNLJAiEFxtuZGYTgX7APyPW9TOznuHzdOBUDt/WcdQqqmrYc6BcvbdFRA4j\nam0W7l5lZtcDrxLcOvuYu68xs1uBJe5eGxyXAQv84NuyJgF/MLMagkC7/XB3UbWF3fvLcIdhqoYS\nEWlUVNss3P0ld5/g7uPc/bZw3U8iggJ3/5m7/6DBfu+7+1R3nxY+PhrNcmqGPBGJdNZZZx3Swe6e\ne+7h2muvPew+ffoEd/pnZ2dz6aWXNrrNmWeeSXO3+N9zzz2UlJTUvZ4zZw4FBQUtLXrUdJQG7pjK\nKaztva0rCxGBefPmsWDBgoPWLViwgHnz5jW779ChQ/nrX/96xJ/dMCxeeukl0tLSjvh4bUVjQ1E/\nQ56uLEQ6oG99C8K5JdrM9Olwzz2HffvSSy/l5ptvpqKigsTERLZt20Z2djYzZszg7LPPZt++fVRW\nVvKLX/yCuXPnHrTvtm3b+Jd/+RdWr15NaWkpV199NStXrmTixImUlpbWbXfttdeyePFiSktLufTS\nS7nlllu47777yM7O5qyzziI9PZ2FCxcyevRolixZQnp6OnfffXfdiLVf//rX+da3vsW2bdu44IIL\nmD17Nu+//z7Dhg3jhRdeIDm5bb/PdGVBUA2VmpxA757KThGB/v37M2vWLF5++WUguKr44he/SHJy\nMs899xzLli1j4cKFfOc736GpUTB+97vf0atXL9atW8ctt9xyUH+J2267jSVLlrBq1SrefvttVq1a\nxY033sjQoUNZuHAhCxcuPOhYS5cu5fHHH+fDDz/kgw8+4OGHH64brnzjxo1cd911rFmzhrS0NJ59\n9tk2/5no25FgEEGNCSXSQTVxBRBNtVVRc+fOZcGCBTz66KO4Oz/60Y945513iIuLIysri927dzN4\n8OBGj/HOO+9w4403AnD88cdz/PHH1733zDPP8NBDD1FVVUVOTg5r16496P2G3nvvPS6++OK6UW8v\nueQS3n33XS666CLGjBlTNxlS5PDmbUlXFkB2YRnDdNusiESYO3cub7zxBsuWLaOkpISZM2fy5JNP\nkpuby9KlS1mxYgWDBg1qdEjy5mzdupVf//rXvPHGG6xatYoLL7zwiI5Tq3Zoc4je8OYKC4JqKPXe\nFpFIffr04ayzzuKaa66pa9guLCxk4MCBJCQksHDhQj755JMmj3H66afz1FNPAbB69WpWrVoFBEOb\n9+7dm9TUVHbv3l1X3QWQkpLCgQMHDjnWaaedxvPPP09JSQnFxcU899xznHbaaW11us3q9tVQxeVV\nFJZWah4LETnEvHnzuPjii+vujPryl7/MZz/7WaZOnUpmZiYTJ05scv9rr72Wq6++mkmTJjFp0iRm\nzpwJBDPezZgxg4kTJzJixIiDhjafP38+559/fl3bRa0TTjiBq666ilmzZgFBA/eMGTOiUuXUmKgN\nUd7ejnSI8vziCn724hq+kDmc08Yf3fhSItI2NER5dHTUIco7hf69E7lvXlSHnRIR6fTUZiEiIs1S\nWIhIh9RVqsg7iqP9eSosRKTDSUpKYu/evQqMNuLu7N27l6SkI7/rs9u3WYhIxzN8+HB27txJbm5u\nrIvSZSQlJTF8+PAj3l9hISIdTkJCAmPGjIl1MSSCqqFERKRZCgsREWmWwkJERJrVZXpwm1ku0PRA\nLU1LB/LaqDgdQVc7H+h659TVzge63jl1tfOBQ89plLs3O3xFlwmLo2VmS1rS5b2z6GrnA13vnLra\n+UDXO6eudj5w5OekaigREWmWwkJERJqlsKj3UKwL0Ma62vlA1zunrnY+0PXOqaudDxzhOanNQkRE\nmqUrCxERaZbCQkREmtXtw8LMzjez9Wa2ycx+EOvytAUz22ZmH5nZCjNr/fSBMWZmj5nZHjNbHbGu\nv5m9bmYbw8d+sSxjax3mnH5mZlnh72mFmc2JZRlbw8xGmNlCM1trZmvM7Jvh+k75e2rifDrz7yjJ\nzBaZ2crwnG4J148xsw/D77z/NrPEFh2vO7dZmFk8sAH4DLATWAzMc/e1MS3YUTKzbUCmu3fKzkRm\ndjpQBPzJ3Y8L190B5Lv77WGo93P378eynK1xmHP6GVDk7r+OZdmOhJkNAYa4+zIzSwGWAp8DrqIT\n/p6aOJ8v0nl/Rwb0dvciM0sA3gO+Cfw78D/uvsDMfg+sdPffNXe87n5lMQvY5O5b3L0CWADMjXGZ\nuj13fwfIb7B6LvBE+PwJgv/IncZhzqnTcvccd18WPj8ArAOG0Ul/T02cT6flgaLwZUK4OPBp4K/h\n+hb/jrp7WAwDdkS83kkn/wcScuA1M1tqZvNjXZg2Msjdc8Lnu4BBsSxMG7rezFaF1VSdosqmITMb\nDcwAPqQL/J4anA904t+RmcWb2QpgD/A6sBkocPeqcJMWf+d197Doqma7+wnABcB1YRVIl+FB3WlX\nqD/9HTAOmA7kAHfFtjitZ2Z9gGeBb7n7/sj3OuPvqZHz6dS/I3evdvfpwHCCmpSJR3qs7h4WWcCI\niNfDw3WdmrtnhY97gOcI/pF0drvDeuXa+uU9MS7PUXP33eF/5hrgYTrZ7ymsB38WeNLd/ydc3Wl/\nT42dT2f/HdVy9wJgIXAykGZmtRPftfg7r7uHxWJgfHh3QCJwGfBijMt0VMysd9hAh5n1Bs4FVje9\nV6fwInBl+PxK4IUYlqVN1H6phi6mE/2ewsbTR4F17n53xFud8vd0uPPp5L+jDDNLC58nE9zIs44g\nNC4NN2vx76hb3w0FEN4Kdw8QDzzm7rfFuEhHxczGElxNQDBt7lOd7ZzM7GngTIKhlHcDPwWeB54B\nRhIMRf9Fd+80DcaHOaczCao3HNgG/FtEfX+HZmazgXeBj4CacPWPCOr5O93vqYnzmUfn/R0dT9CA\nHU9wYfCMu98afkcsAPoDy4GvuHt5s8fr7mEhIiLN6+7VUCIi0gIKCxERaZbCQkREmqWwEBGRZiks\nRESkWQoLkWaYWXXEqKMr2nJ0YjMbHTkSrUhH1aP5TUS6vdJwyASRbktXFiJHKJw35I5w7pBFZnZM\nuH60mb0ZDj73hpmNDNcPMrPnwvkFVprZKeGh4s3s4XDOgdfC3raY2Y3h/AqrzGxBjE5TBFBYiLRE\ncoNqqC9FvFfo7lOBBwhGAgC4H3jC3Y8HngTuC9ffB7zt7tOAE4A14frxwIPuPgUoAD4frv8BMCM8\nzjeidXIiLaEe3CLNMLMid+/TyPptwKfdfUs4CN0udx9gZnkEE+lUhutz3D3dzHKB4ZFDK4TDYb/u\n7uPD198HEtz9F2b2CsGESc8Dz0fMTSDS7nRlIXJ0/DDPWyNyXJ5q6tsSLwQeJLgKWRwxUqhIu1NY\niBydL0U8/jN8/j7BCMYAXyYYoA7gDeBaqJuUJvVwBzWzOGCEuy8Evg+kAodc3Yi0F/2lItK85HC2\nsVqvuHvt7bP9zGwVwdXBvHDdDcDjZvZdIBe4Olz/TeAhM/sawRXEtQQT6jQmHvivMFAMuC+ck0Ak\nJtRmIXKEwjaLTHfPi3VZRKJN1VAiItIsXVmIiEizdGUhIiLNUliIiEizFBYiItIshYWIiDRLYSEi\nIs36/2/TdaG6pdxsAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2a9f244f98>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "mm = training_time // 60\n",
    "ss = training_time % 60\n",
    "print('Training {} epochs in {}:{}'.format(epochs, int(mm), round(ss, 1)))\n",
    "\n",
    "# show the loss and accuracy\n",
    "loss = training.history['loss']\n",
    "val_loss = training.history['val_loss']\n",
    "acc = training.history['acc']\n",
    "val_acc = training.history['val_acc']\n",
    "\n",
    "# loss plot\n",
    "tra = plt.plot(loss)\n",
    "val = plt.plot(val_loss, 'r')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Loss')\n",
    "plt.title('Loss')\n",
    "plt.legend([\"Training\", \"Validation\"])\n",
    "\n",
    "plt.show()\n",
    "\n",
    "# accuracy plot\n",
    "plt.plot(acc)\n",
    "plt.plot(val_acc, 'r')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Loss')\n",
    "plt.title('Accuracy')\n",
    "plt.legend(['Training', 'Validation'], loc=4)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "_cell_guid": "76748dc5-cd6c-4af9-b4da-6c7c34baa75a",
    "_uuid": "391704a5b8526a35012390e2a4b080014ec93ccf",
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "test_images = test_data.iloc[:, 1:]\n",
    "test_labels = test_data.iloc[:, 0]\n",
    "test_labels = test_labels.as_matrix()\n",
    "test_features = test_images.as_matrix().reshape(test_images.shape[0], 1, 28, 28)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "_cell_guid": "95d2c70b-6d36-43a5-9e87-fba511e5e203",
    "_uuid": "da3c749102e096c5a27cba30fbb401c1a4f02757",
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import accuracy_score\n",
    "pred = model.predict(test_features)\n",
    "# convert predicions from categorical back to 0...9 digits\n",
    "pred_digits = np.argmax(pred, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "_cell_guid": "6eeec6de-125a-42cf-aaf2-7d1fc5a97bdc",
    "_uuid": "7f8c2336391815120ff1dca3afb9b46636038497"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.93230000000000002"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(test_labels, pred_digits)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_cell_guid": "779e52c2-ffeb-4a19-a918-0ac3398548fe",
    "_uuid": "d09074914146365c5f3575f81033592fbb3622e3",
    "collapsed": true
   },
   "source": [
    "# original MNIST"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://s3.amazonaws.com/img-datasets/mnist.npz\n",
      "11460608/11490434 [============================>.] - ETA: 0s"
     ]
    }
   ],
   "source": [
    "from keras.datasets import mnist\n",
    "\n",
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAADuCAYAAAAp6fzCAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ni4jkxP8HHtH0+tf6By0AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2a3d6a1be0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "f, axes = plt.subplots(2, 4)\n",
    "for row in axes:\n",
    "    for axe in row:\n",
    "        index = np.random.randint(10000)\n",
    "        img = x_train[index]\n",
    "        obj = y_train[index]\n",
    "        axe.imshow(img, cmap='gray')\n",
    "        axe.set_title(obj)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "train_features = x_train.reshape((x_train.shape[0], 1, 28, 28))\n",
    "test_features = x_test.reshape((x_test.shape[0], 1, 28, 28))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train_labels = np_utils.to_categorical(y_train)\n",
    "\n",
    "# divide data into training and validation set\n",
    "train_features, vali_features, train_labels, vali_labels = train_test_split(train_features, train_labels, \n",
    "                                                                           test_size=0.15, random_state=80)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# hyperparameters\n",
    "epochs = 30\n",
    "batch_size = 256\n",
    "\n",
    "# build model2\n",
    "n_classes = 10\n",
    "model2 = Sequential()\n",
    "\n",
    "\n",
    "# input layer\n",
    "model2.add(InputLayer(input_shape=(1, 28, 28)))\n",
    "\n",
    "# normalization\n",
    "model2.add(BatchNormalization())\n",
    "\n",
    "model2.add(Conv2D(64, (5, 5), activation='relu',\n",
    "          bias_initializer='RandomNormal', kernel_initializer='random_uniform'))\n",
    "\n",
    "\n",
    "model2.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "model2.add(Conv2D(512, (5, 5), activation='relu'))\n",
    "\n",
    "model2.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "model2.add(Flatten())\n",
    "\n",
    "model2.add(Dense(128, activation='relu'))\n",
    "\n",
    "model2.add(Dropout(0.35))\n",
    "\n",
    "model2.add(Dense(64, activation='relu'))\n",
    "\n",
    "model2.add(Dropout(0.35))\n",
    "\n",
    "model2.add(Dense(n_classes, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "opt = Adam(decay=0.001)\n",
    "# compile the model\n",
    "model2.compile(optimizer=opt,\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 51000 samples, validate on 9000 samples\n",
      "Epoch 1/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.3541 - acc: 0.8907 - val_loss: 0.0734 - val_acc: 0.9794\n",
      "Epoch 2/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0910 - acc: 0.9757 - val_loss: 0.0436 - val_acc: 0.9872\n",
      "Epoch 3/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0623 - acc: 0.9838 - val_loss: 0.0435 - val_acc: 0.9878\n",
      "Epoch 4/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0468 - acc: 0.9871 - val_loss: 0.0331 - val_acc: 0.9901\n",
      "Epoch 5/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0383 - acc: 0.9902 - val_loss: 0.0301 - val_acc: 0.9916\n",
      "Epoch 6/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0264 - acc: 0.9925 - val_loss: 0.0340 - val_acc: 0.9916\n",
      "Epoch 7/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0238 - acc: 0.9935 - val_loss: 0.0294 - val_acc: 0.9923\n",
      "Epoch 8/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0198 - acc: 0.9947 - val_loss: 0.0316 - val_acc: 0.9926\n",
      "Epoch 9/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0160 - acc: 0.9954 - val_loss: 0.0313 - val_acc: 0.9933\n",
      "Epoch 10/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0146 - acc: 0.9960 - val_loss: 0.0301 - val_acc: 0.9930\n",
      "Epoch 11/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0117 - acc: 0.9970 - val_loss: 0.0325 - val_acc: 0.9931\n",
      "Epoch 12/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0111 - acc: 0.9967 - val_loss: 0.0319 - val_acc: 0.9933\n",
      "Epoch 13/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0111 - acc: 0.9971 - val_loss: 0.0306 - val_acc: 0.9936\n",
      "Epoch 14/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0095 - acc: 0.9974 - val_loss: 0.0308 - val_acc: 0.9938\n",
      "Epoch 15/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0073 - acc: 0.9982 - val_loss: 0.0329 - val_acc: 0.9930\n",
      "Epoch 16/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0066 - acc: 0.9981 - val_loss: 0.0347 - val_acc: 0.9927\n",
      "Epoch 17/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0062 - acc: 0.9982 - val_loss: 0.0310 - val_acc: 0.9946\n",
      "Epoch 18/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0049 - acc: 0.9989 - val_loss: 0.0367 - val_acc: 0.9932\n",
      "Epoch 19/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0051 - acc: 0.9984 - val_loss: 0.0344 - val_acc: 0.9939\n",
      "Epoch 20/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0052 - acc: 0.9985 - val_loss: 0.0376 - val_acc: 0.9933\n",
      "Epoch 21/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0044 - acc: 0.9987 - val_loss: 0.0359 - val_acc: 0.9943\n",
      "Epoch 22/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0038 - acc: 0.9988 - val_loss: 0.0334 - val_acc: 0.9942\n",
      "Epoch 23/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0035 - acc: 0.9988 - val_loss: 0.0351 - val_acc: 0.9939\n",
      "Epoch 24/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0042 - acc: 0.9987 - val_loss: 0.0312 - val_acc: 0.9946\n",
      "Epoch 25/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0042 - acc: 0.9989 - val_loss: 0.0386 - val_acc: 0.9932\n",
      "Epoch 26/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0042 - acc: 0.9988 - val_loss: 0.0344 - val_acc: 0.9937\n",
      "Epoch 27/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0026 - acc: 0.9993 - val_loss: 0.0375 - val_acc: 0.9938\n",
      "Epoch 28/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0028 - acc: 0.9994 - val_loss: 0.0358 - val_acc: 0.9943\n",
      "Epoch 29/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0020 - acc: 0.9995 - val_loss: 0.0373 - val_acc: 0.9940\n",
      "Epoch 30/30\n",
      "51000/51000 [==============================] - 24s - loss: 0.0021 - acc: 0.9994 - val_loss: 0.0419 - val_acc: 0.9943\n"
     ]
    }
   ],
   "source": [
    "training2 = model2.fit(train_features, train_labels,\n",
    "                     validation_data=(vali_features, vali_labels),\n",
    "                     epochs=epochs,\n",
    "                     batch_size=batch_size, \n",
    "                     verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training 30 epochs in 12:11.4\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYsAAAEWCAYAAACXGLsWAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2a6ed51438>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYsAAAEWCAYAAACXGLsWAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSFr/H/U4Y0f2fdihAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f2a3d703d68>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "mm = training_time // 60\n",
    "ss = training_time % 60\n",
    "print('Training {} epochs in {}:{}'.format(epochs, int(mm), round(ss, 1)))\n",
    "\n",
    "# show the loss and accuracy\n",
    "loss = training2.history['loss']\n",
    "val_loss = training2.history['val_loss']\n",
    "acc = training2.history['acc']\n",
    "val_acc = training2.history['val_acc']\n",
    "\n",
    "# loss plot\n",
    "tra = plt.plot(loss)\n",
    "val = plt.plot(val_loss, 'r')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Loss')\n",
    "plt.title('Loss')\n",
    "plt.legend([\"Training\", \"Validation\"])\n",
    "\n",
    "plt.show()\n",
    "\n",
    "# accuracy plot\n",
    "plt.plot(acc)\n",
    "plt.plot(val_acc, 'r')\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Loss')\n",
    "plt.title('Accuracy')\n",
    "plt.legend(['Training', 'Validation'], loc=4)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_labels = y_test\n",
    "pred = model2.predict(test_features)\n",
    "# convert predicions from categorical back to 0...9 digits\n",
    "pred_digits = np.argmax(pred, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.99419999999999997"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(test_labels, pred_digits)"
   ]
  },
  {
   "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": "3.5.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}