13.5 Quick Draw Autoencoder.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.layers import Input, Dense, Conv2D, MaxPooling2D, UpSampling2D, Reshape, Concatenate\n",
    "from keras.models import Model\n",
    "from keras.preprocessing.image import load_img, img_to_array, ImageDataGenerator\n",
    "from keras import backend as K\n",
    "from struct import unpack\n",
    "import matplotlib.pyplot as plt\n",
    "import json\n",
    "import glob\n",
    "from keras.datasets import mnist\n",
    "from sklearn.model_selection import train_test_split\n",
    "import numpy as np\n",
    "from io import BytesIO\n",
    "import PIL\n",
    "from PIL import ImageDraw\n",
    "from keras.utils import get_file\n",
    "from IPython.display import clear_output, Image, display, HTML"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'camel': '/home/douwe/.keras/datasets/camel.bin',\n",
       " 'cat': '/home/douwe/.keras/datasets/cat.bin',\n",
       " 'cow': '/home/douwe/.keras/datasets/cow.bin',\n",
       " 'dog': '/home/douwe/.keras/datasets/dog.bin',\n",
       " 'horse': '/home/douwe/.keras/datasets/horse.bin',\n",
       " 'lion': '/home/douwe/.keras/datasets/lion.bin',\n",
       " 'mouse': '/home/douwe/.keras/datasets/mouse.bin',\n",
       " 'pig': '/home/douwe/.keras/datasets/pig.bin',\n",
       " 'rabbit': '/home/douwe/.keras/datasets/rabbit.bin',\n",
       " 'sheep': '/home/douwe/.keras/datasets/sheep.bin',\n",
       " 'tiger': '/home/douwe/.keras/datasets/tiger.bin',\n",
       " 'zebra': '/home/douwe/.keras/datasets/zebra.bin'}"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels = ['camel', 'cat', 'cow', 'dog', 'horse', 'lion', 'mouse', 'pig', 'rabbit', 'sheep', 'tiger', 'zebra']\n",
    "labels_to_idx = {label: idx for idx, label in enumerate(labels)}\n",
    "base_url = 'https://storage.googleapis.com/quickdraw_dataset/full/binary/%s.bin'\n",
    "paths = {label: get_file(label + '.bin', base_url % label) for label in labels}\n",
    "paths"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((102000, 32, 32, 1), (18000, 32, 32, 1))"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def load_icons(paths, train_size=0.85, max_per_class=10000):\n",
    "    x = []\n",
    "    y = []\n",
    "    for label, path in paths.items():\n",
    "        with open(path, 'rb') as f:\n",
    "            while True:\n",
    "                img = PIL.Image.new('L', (32, 32), 'white')\n",
    "                draw = ImageDraw.Draw(img)\n",
    "                header = f.read(15)\n",
    "                if len(header) != 15:\n",
    "                    break\n",
    "                strokes, = unpack('H', f.read(2))\n",
    "                for i in range(strokes):\n",
    "                    n_points, = unpack('H', f.read(2))\n",
    "                    fmt = str(n_points) + 'B'\n",
    "                    read_scaled = lambda: (p // 8 for \n",
    "                                           p in unpack(fmt, f.read(n_points)))\n",
    "                    points = [*zip(read_scaled(), read_scaled())]\n",
    "                    draw.line(points, fill=0)\n",
    "                img = img_to_array(img)\n",
    "                x.append(img)\n",
    "                y.append(labels_to_idx[label])\n",
    "                if len(x) % max_per_class == 0:\n",
    "                    break\n",
    "    x = np.asarray(x) / 255\n",
    "    y = np.asarray(y)\n",
    "    return train_test_split(x, y, train_size=train_size)\n",
    "\n",
    "\n",
    "x_train, x_test, y_train, y_test = load_icons(paths)\n",
    "x_train.shape, x_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "____________________________________________________________________________________________________\n",
      "Layer (type)                     Output Shape          Param #     Connected to                     \n",
      "====================================================================================================\n",
      "input_1 (InputLayer)             (None, 32, 32, 1)     0                                            \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_1 (Conv2D)                (None, 32, 32, 4)     40                                           \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_2 (Conv2D)                (None, 32, 32, 4)     20                                           \n",
      "____________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)      (None, 32, 32, 8)     0                                            \n",
      "____________________________________________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2D)   (None, 16, 16, 8)     0                                            \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_3 (Conv2D)                (None, 16, 16, 8)     584                                          \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_4 (Conv2D)                (None, 16, 16, 8)     264                                          \n",
      "____________________________________________________________________________________________________\n",
      "concatenate_2 (Concatenate)      (None, 16, 16, 16)    0                                            \n",
      "____________________________________________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2D)   (None, 8, 8, 16)      0                                            \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_5 (Conv2D)                (None, 8, 8, 16)      2320                                         \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_6 (Conv2D)                (None, 8, 8, 16)      1040                                         \n",
      "____________________________________________________________________________________________________\n",
      "concatenate_3 (Concatenate)      (None, 8, 8, 32)      0                                            \n",
      "____________________________________________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2D)   (None, 4, 4, 32)      0                                            \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_7 (Conv2D)                (None, 4, 4, 32)      9248                                         \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_8 (Conv2D)                (None, 4, 4, 32)      4128                                         \n",
      "____________________________________________________________________________________________________\n",
      "concatenate_4 (Concatenate)      (None, 4, 4, 64)      0                                            \n",
      "____________________________________________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2D)   (None, 2, 2, 64)      0                                            \n",
      "____________________________________________________________________________________________________\n",
      "dense_1 (Dense)                  (None, 2, 2, 32)      2080                                         \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_9 (Conv2D)                (None, 2, 2, 32)      9248                                         \n",
      "____________________________________________________________________________________________________\n",
      "up_sampling2d_1 (UpSampling2D)   (None, 4, 4, 32)      0                                            \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_10 (Conv2D)               (None, 4, 4, 16)      4624                                         \n",
      "____________________________________________________________________________________________________\n",
      "up_sampling2d_2 (UpSampling2D)   (None, 8, 8, 16)      0                                            \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_11 (Conv2D)               (None, 8, 8, 8)       1160                                         \n",
      "____________________________________________________________________________________________________\n",
      "up_sampling2d_3 (UpSampling2D)   (None, 16, 16, 8)     0                                            \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_12 (Conv2D)               (None, 16, 16, 4)     292                                          \n",
      "____________________________________________________________________________________________________\n",
      "up_sampling2d_4 (UpSampling2D)   (None, 32, 32, 4)     0                                            \n",
      "____________________________________________________________________________________________________\n",
      "conv2d_13 (Conv2D)               (None, 32, 32, 1)     37                                           \n",
      "====================================================================================================\n",
      "Total params: 35,085\n",
      "Trainable params: 35,085\n",
      "Non-trainable params: 0\n",
      "____________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "def create_autoencoder():\n",
    "    input_img = Input(shape=(32, 32, 1))\n",
    "\n",
    "    channels = 4\n",
    "    x = input_img\n",
    "    for i in range(4):\n",
    "        left = Conv2D(channels, (3, 3), activation='relu', padding='same')(x)\n",
    "        right = Conv2D(channels, (2, 2), activation='relu', padding='same')(x)\n",
    "        conc = Concatenate()([left, right])\n",
    "        x = MaxPooling2D((2, 2), padding='same')(conc)\n",
    "        channels *= 2\n",
    "\n",
    "    x = Dense(32)(x)\n",
    "\n",
    "    # at this point the representation is (4, 4, 8) i.e. 128-dimensional\n",
    "\n",
    "    x = Conv2D(32, (3, 3), activation='relu', padding='same')(x)\n",
    "    x = UpSampling2D((2, 2))(x)\n",
    "    x = Conv2D(16, (3, 3), activation='relu', padding='same')(x)\n",
    "    x = UpSampling2D((2, 2))(x)\n",
    "    x = Conv2D(8, (3, 3), activation='relu', padding='same')(x)\n",
    "    x = UpSampling2D((2, 2))(x)\n",
    "    x = Conv2D(4, (3, 3), activation='relu', padding='same')(x)\n",
    "    x = UpSampling2D((2, 2))(x)\n",
    "    decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)\n",
    "\n",
    "    autoencoder = Model(input_img, decoded)\n",
    "    autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')\n",
    "    return autoencoder\n",
    "\n",
    "autoencoder = create_autoencoder()\n",
    "autoencoder.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 102000 samples, validate on 18000 samples\n",
      "Epoch 1/100\n",
      "102000/102000 [==============================] - 10s - loss: 0.6226 - val_loss: 0.5540\n",
      "Epoch 2/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.5006 - val_loss: 0.4565\n",
      "Epoch 3/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.4300 - val_loss: 0.4106\n",
      "Epoch 4/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.4004 - val_loss: 0.3816\n",
      "Epoch 5/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.3638 - val_loss: 0.3481\n",
      "Epoch 6/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.3388 - val_loss: 0.3284\n",
      "Epoch 7/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.3218 - val_loss: 0.3150\n",
      "Epoch 8/100\n",
      "102000/102000 [==============================] - 8s - loss: 0.3096 - val_loss: 0.3051\n",
      "Epoch 9/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.3001 - val_loss: 0.2970\n",
      "Epoch 10/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2927 - val_loss: 0.2884\n",
      "Epoch 11/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2867 - val_loss: 0.2835\n",
      "Epoch 12/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2819 - val_loss: 0.2780\n",
      "Epoch 13/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2779 - val_loss: 0.2761\n",
      "Epoch 14/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2743 - val_loss: 0.2725\n",
      "Epoch 15/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2712 - val_loss: 0.2701\n",
      "Epoch 16/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2687 - val_loss: 0.2666\n",
      "Epoch 17/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2661 - val_loss: 0.2646\n",
      "Epoch 18/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2640 - val_loss: 0.2642\n",
      "Epoch 19/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2617 - val_loss: 0.2615\n",
      "Epoch 20/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2602 - val_loss: 0.2585\n",
      "Epoch 21/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2585 - val_loss: 0.2565\n",
      "Epoch 22/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2572 - val_loss: 0.2562\n",
      "Epoch 23/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2558 - val_loss: 0.2554\n",
      "Epoch 24/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2547 - val_loss: 0.2535\n",
      "Epoch 25/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2535 - val_loss: 0.2543\n",
      "Epoch 26/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2524 - val_loss: 0.2524\n",
      "Epoch 27/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2513 - val_loss: 0.2503\n",
      "Epoch 28/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2504 - val_loss: 0.2484\n",
      "Epoch 29/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2495 - val_loss: 0.2473\n",
      "Epoch 30/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2485 - val_loss: 0.2490\n",
      "Epoch 31/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2478 - val_loss: 0.2461\n",
      "Epoch 32/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2472 - val_loss: 0.2446\n",
      "Epoch 33/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2464 - val_loss: 0.2469\n",
      "Epoch 34/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2460 - val_loss: 0.2468\n",
      "Epoch 35/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2452 - val_loss: 0.2451\n",
      "Epoch 36/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2446 - val_loss: 0.2435\n",
      "Epoch 37/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2436 - val_loss: 0.2420\n",
      "Epoch 38/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2432 - val_loss: 0.2413\n",
      "Epoch 39/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2427 - val_loss: 0.2407\n",
      "Epoch 40/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2422 - val_loss: 0.2404\n",
      "Epoch 41/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2418 - val_loss: 0.2410\n",
      "Epoch 42/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2412 - val_loss: 0.2418\n",
      "Epoch 43/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2407 - val_loss: 0.2413\n",
      "Epoch 44/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2402 - val_loss: 0.2390\n",
      "Epoch 45/100\n",
      "102000/102000 [==============================] - 8s - loss: 0.2395 - val_loss: 0.2375\n",
      "Epoch 46/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2395 - val_loss: 0.2401\n",
      "Epoch 47/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2392 - val_loss: 0.2393\n",
      "Epoch 48/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2388 - val_loss: 0.2378\n",
      "Epoch 49/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2382 - val_loss: 0.2400\n",
      "Epoch 50/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2377 - val_loss: 0.2359\n",
      "Epoch 51/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2369 - val_loss: 0.2371\n",
      "Epoch 52/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2364 - val_loss: 0.2364\n",
      "Epoch 53/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2362 - val_loss: 0.2358\n",
      "Epoch 54/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2360 - val_loss: 0.2362\n",
      "Epoch 55/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2356 - val_loss: 0.2387\n",
      "Epoch 56/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2352 - val_loss: 0.2346\n",
      "Epoch 57/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2349 - val_loss: 0.2347\n",
      "Epoch 58/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2346 - val_loss: 0.2327\n",
      "Epoch 59/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2341 - val_loss: 0.2351\n",
      "Epoch 60/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2339 - val_loss: 0.2339\n",
      "Epoch 61/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2336 - val_loss: 0.2339\n",
      "Epoch 62/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2332 - val_loss: 0.2338\n",
      "Epoch 63/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2329 - val_loss: 0.2320\n",
      "Epoch 64/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2326 - val_loss: 0.2330\n",
      "Epoch 65/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2323 - val_loss: 0.2340\n",
      "Epoch 66/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2319 - val_loss: 0.2332\n",
      "Epoch 67/100\n",
      "102000/102000 [==============================] - 8s - loss: 0.2317 - val_loss: 0.2356\n",
      "Epoch 68/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2314 - val_loss: 0.2313\n",
      "Epoch 69/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2311 - val_loss: 0.2328\n",
      "Epoch 70/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2311 - val_loss: 0.2292\n",
      "Epoch 71/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2307 - val_loss: 0.2295\n",
      "Epoch 72/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2303 - val_loss: 0.2295\n",
      "Epoch 73/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2299 - val_loss: 0.2276\n",
      "Epoch 74/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2299 - val_loss: 0.2288\n",
      "Epoch 75/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2296 - val_loss: 0.2320\n",
      "Epoch 76/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2295 - val_loss: 0.2302\n",
      "Epoch 77/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2293 - val_loss: 0.2334\n",
      "Epoch 78/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2289 - val_loss: 0.2280\n",
      "Epoch 79/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2288 - val_loss: 0.2304\n",
      "Epoch 80/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2287 - val_loss: 0.2289\n",
      "Epoch 81/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2285 - val_loss: 0.2282\n",
      "Epoch 82/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2280 - val_loss: 0.2283\n",
      "Epoch 83/100\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "102000/102000 [==============================] - 7s - loss: 0.2278 - val_loss: 0.2285\n",
      "Epoch 84/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2276 - val_loss: 0.2267\n",
      "Epoch 85/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2274 - val_loss: 0.2260\n",
      "Epoch 86/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2271 - val_loss: 0.2276\n",
      "Epoch 87/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2272 - val_loss: 0.2274\n",
      "Epoch 88/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2270 - val_loss: 0.2268\n",
      "Epoch 89/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2269 - val_loss: 0.2259\n",
      "Epoch 90/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2265 - val_loss: 0.2279\n",
      "Epoch 91/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2262 - val_loss: 0.2262\n",
      "Epoch 92/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2260 - val_loss: 0.2284\n",
      "Epoch 93/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2257 - val_loss: 0.2261\n",
      "Epoch 94/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2257 - val_loss: 0.2248\n",
      "Epoch 95/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2257 - val_loss: 0.2240\n",
      "Epoch 96/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2255 - val_loss: 0.2285\n",
      "Epoch 97/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2254 - val_loss: 0.2271\n",
      "Epoch 98/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2253 - val_loss: 0.2247\n",
      "Epoch 99/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2248 - val_loss: 0.2245\n",
      "Epoch 100/100\n",
      "102000/102000 [==============================] - 7s - loss: 0.2246 - val_loss: 0.2249\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f625db94d30>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from keras.callbacks import TensorBoard\n",
    "\n",
    "autoencoder.fit(x_train, x_train,\n",
    "                epochs=100,\n",
    "                batch_size=128,\n",
    "                shuffle=True,\n",
    "                validation_data=(x_test, x_test),\n",
    "                callbacks=[TensorBoard(log_dir='/tmp/autoencoder')])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(25, 32, 32, 1)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cols = 25\n",
    "idx = np.random.randint(x_test.shape[0], size=cols)\n",
    "sample = x_test[idx]\n",
    "decoded_imgs = autoencoder.predict(sample)\n",
    "decoded_imgs.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAyAAAABUCAIAAAD1ScR1AAB2LUlEQVR4nO19d3Rc133me9N7b8AM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\nOuqoo441Rt3AqqOOOuqoo4466lhj/F+oBP91mOV3UgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "def decode_img(tile, factor=1.0):\n",
    "    tile = tile.reshape(tile.shape[:-1])\n",
    "    tile = np.clip(tile * 255, 0, 255)\n",
    "    return PIL.Image.fromarray(tile)\n",
    "    \n",
    "\n",
    "overview = PIL.Image.new('RGB', (cols * 32, 64 + 20), (128, 128, 128))\n",
    "for idx in range(cols):\n",
    "    overview.paste(decode_img(sample[idx]), (idx * 32, 5))\n",
    "    overview.paste(decode_img(decoded_imgs[idx]), (idx * 32, 42))\n",
    "f = BytesIO()\n",
    "overview.save(f, 'png')\n",
    "display(Image(data=f.getvalue()))\n",
    "\n"
   ]
  },
  {
   "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}