Day-2/AI_ML_Intro/Autoencoders_with_MNIST_dataset.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "7rFWB1DOtEdC"
   },
   "outputs": [],
   "source": [
    "# Autoencoders data compression algorithm\n",
    "# Applications : Data denoising and dimensionality reduction for data visualization\n",
    "# Requirements :  - an encoding function\n",
    "#                 - a decoding function\n",
    "#                 - a loss function (distance function between the amount of information loss \n",
    "#                between the compressed representation of your data and the decompressed representation)\n",
    "# \n",
    "# This exercise is adapted from https://blog.keras.io/building-autoencoders-in-keras.html by Lavsen Dahal"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "rMnvEJzGtEde"
   },
   "source": [
    "# Single Fully Connected layer as encoder and decoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "8eEeb45BtEdq",
    "outputId": "fffe4bee-3a4d-4a46-f25a-38d740f8865d"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.layers import Input, Dense\n",
    "from keras.models import Model\n",
    "\n",
    "# this is the size of our encoded representations\n",
    "encoding_dim = 32  # 32 floats ->compression factor - 24.5, assuming the input is 784 floats (MNIST data image size- 28x28)\n",
    "\n",
    "# this is our input placeholder\n",
    "input_img = Input(shape=(784,))\n",
    "# \"encoded\" is the encoded representation of the input\n",
    "encoded = Dense(encoding_dim, activation='relu')(input_img)\n",
    "# \"decoded\" is the lossy reconstruction of the input\n",
    "decoded = Dense(784, activation='sigmoid')(encoded)\n",
    "\n",
    "# this model maps an input to its reconstruction\n",
    "autoencoder = Model(input_img, decoded)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "CBqRVrRQtEd-"
   },
   "outputs": [],
   "source": [
    "# this model maps an input to its encoded representation\n",
    "encoder = Model(input_img, encoded)\n",
    "# create a placeholder for an encoded (32-dimensional) input\n",
    "encoded_input = Input(shape=(encoding_dim,))\n",
    "# retrieve the last layer of the autoencoder model\n",
    "decoder_layer = autoencoder.layers[-1]\n",
    "# create the decoder model\n",
    "decoder = Model(encoded_input, decoder_layer(encoded_input))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "u99CyRdxtEeL"
   },
   "outputs": [],
   "source": [
    "autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "9lxDFEtytEeZ"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading data from https://s3.amazonaws.com/img-datasets/mnist.npz\n",
      "11493376/11490434 [==============================] - 234s 20us/step\n"
     ]
    }
   ],
   "source": [
    "#Input data - MNIST\n",
    "from keras.datasets import mnist\n",
    "import numpy as np\n",
    "(x_train, _), (x_test, _) = mnist.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "WvR1qmW3tEep",
    "outputId": "89859cee-536c-4304-b17d-e583ef2ba552"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(60000, 784)\n",
      "(10000, 784)\n"
     ]
    }
   ],
   "source": [
    "# Normalize the images and flatten the array. \n",
    "x_train = x_train.astype('float32') / 255.\n",
    "x_test = x_test.astype('float32') / 255.\n",
    "x_train = x_train.reshape((len(x_train), np.prod(x_train.shape[1:])))\n",
    "x_test = x_test.reshape((len(x_test), np.prod(x_test.shape[1:])))\n",
    "print (x_train.shape)\n",
    "print (x_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "22wwesCDtEe1",
    "outputId": "b4ef40f2-fefe-44c0-beb5-726528b283c9"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 60000 samples, validate on 10000 samples\n",
      "Epoch 1/50\n",
      "60000/60000 [==============================] - 3s 51us/step - loss: 0.3619 - val_loss: 0.2710\n",
      "Epoch 2/50\n",
      "60000/60000 [==============================] - 1s 14us/step - loss: 0.2630 - val_loss: 0.2513\n",
      "Epoch 3/50\n",
      "60000/60000 [==============================] - 1s 14us/step - loss: 0.2412 - val_loss: 0.2287\n",
      "Epoch 4/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.2212 - val_loss: 0.2109\n",
      "Epoch 5/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.2059 - val_loss: 0.1982\n",
      "Epoch 6/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1950 - val_loss: 0.1892\n",
      "Epoch 7/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1869 - val_loss: 0.1820\n",
      "Epoch 8/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1804 - val_loss: 0.1761\n",
      "Epoch 9/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1749 - val_loss: 0.1708\n",
      "Epoch 10/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1699 - val_loss: 0.1662\n",
      "Epoch 11/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1655 - val_loss: 0.1619\n",
      "Epoch 12/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1613 - val_loss: 0.1580\n",
      "Epoch 13/50\n",
      "60000/60000 [==============================] - 1s 14us/step - loss: 0.1575 - val_loss: 0.1542\n",
      "Epoch 14/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1539 - val_loss: 0.1508\n",
      "Epoch 15/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1506 - val_loss: 0.1475\n",
      "Epoch 16/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1476 - val_loss: 0.1448\n",
      "Epoch 17/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1448 - val_loss: 0.1420\n",
      "Epoch 18/50\n",
      "60000/60000 [==============================] - 1s 16us/step - loss: 0.1423 - val_loss: 0.1396\n",
      "Epoch 19/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1399 - val_loss: 0.1373\n",
      "Epoch 20/50\n",
      "60000/60000 [==============================] - 1s 14us/step - loss: 0.1378 - val_loss: 0.1352\n",
      "Epoch 21/50\n",
      "60000/60000 [==============================] - 1s 17us/step - loss: 0.1357 - val_loss: 0.1331\n",
      "Epoch 22/50\n",
      "60000/60000 [==============================] - 1s 16us/step - loss: 0.1337 - val_loss: 0.1312\n",
      "Epoch 23/50\n",
      "60000/60000 [==============================] - 1s 14us/step - loss: 0.1319 - val_loss: 0.1294\n",
      "Epoch 24/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1301 - val_loss: 0.1276\n",
      "Epoch 25/50\n",
      "60000/60000 [==============================] - 1s 14us/step - loss: 0.1283 - val_loss: 0.1259\n",
      "Epoch 26/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1266 - val_loss: 0.1242\n",
      "Epoch 27/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1250 - val_loss: 0.1226\n",
      "Epoch 28/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1235 - val_loss: 0.1212\n",
      "Epoch 29/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1221 - val_loss: 0.1198\n",
      "Epoch 30/50\n",
      "60000/60000 [==============================] - 1s 14us/step - loss: 0.1207 - val_loss: 0.1185\n",
      "Epoch 31/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1195 - val_loss: 0.1173\n",
      "Epoch 32/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1183 - val_loss: 0.1161\n",
      "Epoch 33/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1172 - val_loss: 0.1150\n",
      "Epoch 34/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1161 - val_loss: 0.1140\n",
      "Epoch 35/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1152 - val_loss: 0.1131\n",
      "Epoch 36/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1143 - val_loss: 0.1122\n",
      "Epoch 37/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1135 - val_loss: 0.1114\n",
      "Epoch 38/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1127 - val_loss: 0.1107\n",
      "Epoch 39/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1120 - val_loss: 0.1100\n",
      "Epoch 40/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1114 - val_loss: 0.1094\n",
      "Epoch 41/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1108 - val_loss: 0.1088\n",
      "Epoch 42/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1102 - val_loss: 0.1083\n",
      "Epoch 43/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1097 - val_loss: 0.1078\n",
      "Epoch 44/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1092 - val_loss: 0.1073\n",
      "Epoch 45/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1088 - val_loss: 0.1069\n",
      "Epoch 46/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1084 - val_loss: 0.1065\n",
      "Epoch 47/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1080 - val_loss: 0.1062\n",
      "Epoch 48/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1077 - val_loss: 0.1059\n",
      "Epoch 49/50\n",
      "60000/60000 [==============================] - 1s 12us/step - loss: 0.1074 - val_loss: 0.1055\n",
      "Epoch 50/50\n",
      "60000/60000 [==============================] - 1s 13us/step - loss: 0.1071 - val_loss: 0.1052\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x2b43d56c518>"
      ]
     },
     "execution_count": 7,
     "metadata": {
      "tags": []
     },
     "output_type": "execute_result"
    }
   ],
   "source": [
    "autoencoder.fit(x_train, x_train,\n",
    "                epochs=50,\n",
    "                batch_size=256,\n",
    "                shuffle=True,\n",
    "                validation_data=(x_test, x_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "HLquM4DCtEfE"
   },
   "outputs": [],
   "source": [
    "# encode and decode some digits\n",
    "# note that we take them from the *test* set\n",
    "encoded_imgs = encoder.predict(x_test)\n",
    "decoded_imgs = decoder.predict(encoded_imgs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 0,
   "metadata": {
    "colab": {},
    "colab_type": "code",
    "id": "P9pVBrJBtEfW",
    "outputId": "380fcf0b-2df4-4f3d-cb91-f7d9884ef10b"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 20 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "n = 10  # how many digits we will displaynepa\n",
    "plt.figure(figsize=(20, 4))\n",
    "for i in range(n):\n",
    "    # display original\n",
    "    ax = plt.subplot(2, n, i + 1)\n",
    "    plt.imshow(x_test[i].reshape(28, 28))\n",
    "    plt.gray()\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "\n",
    "    # display reconstruction\n",
    "    ax = plt.subplot(2, n, i + 1 + n)\n",
    "    plt.imshow(decoded_imgs[i].reshape(28, 28))\n",
    "    plt.gray()\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "name": "Autoencoders with MNIST dataset.ipynb",
   "provenance": []
  },
  "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": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}