cap05/C05_E1_Shallow-AutoEncoder.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<div style=\"width: 100%; clear: both;\">\n",
    "<div style=\"float: left; width: 50%;\">\n",
    "<img src=\"http://www.uoc.edu/portal/_resources/common/imatges/marca_UOC/UOC_Masterbrand.jpg\", align=\"left\">\n",
    "</div>\n",
    "<div style=\"float: right; width: 50%;\">\n",
    "<p style=\"margin: 0; padding-top: 22px; text-align:right;\"><a href=\"https://www.editorialuoc.cat/deep-learning\" target=\"_blank\\\">Deep learning: Principios y fundamentos</a></p>\n",
    "<p style=\"margin: 0; text-align:right;\">Estudios de Informática, Multimedia y Telecomunicación</p>\n",
    "<p style=\"margin: 0; text-align:right; padding-button: 100px;\">Universitat Oberta de Catalunya (UOC)</p>\n",
    "</div>\n",
    "</div>\n",
    "<div style=\"width:100%;\">&nbsp;</div>\n",
    "\n",
    "\n",
    "# Capítulo 5: Ejemplo de funcionamiento de Autoencoder"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "En este ejemplo veremos un caso de aplicación para un autoencoder. En concreto, nos centraremos en la compresión y reconstrucción de imágenes empleando *autoencoders*.\n",
    "\n",
    "<u>Nota</u>: Este ejemplo está basado en https://blog.keras.io/building-autoencoders-in-keras.html"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Deshabilitamos la aparición de warnings.\n",
    "\n",
    "<u>Nota</u>: No se recomienda este paso cuando se está desarrollando el código."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "\n",
    "warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Carga del conjunto de datos\n",
    "\n",
    "En primer lugar, cargamos el conjunto de datos, que en este caso será el dataset de dígitos [MNIST](https://en.wikipedia.org/wiki/MNIST_database). \n",
    "\n",
    "Realizamos esta carga, directamente, a partir de la librería [Keras](https://keras.io/)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.datasets import mnist\n",
    "import numpy as np\n",
    "\n",
    "(x_train, _), (x_test, _) = mnist.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(60000, 784)\n",
      "(10000, 784)\n"
     ]
    }
   ],
   "source": [
    "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": "markdown",
   "metadata": {},
   "source": [
    "## 2. Creación del *autoencoder*\n",
    "\n",
    "A continuación, creamos el *autoencoder* en su forma más senzilla:\n",
    "- A partir de las imágenes de entrada de de 28x28 píxeles en escala de grises, se crea una capa con 784 valores de entrada\n",
    "- Creamos un **codificador** (*encoder*) con una representación de 32 neuronas\n",
    "- Creamos un **decodificador** (*decoder*) que genere una salida en la misma resolución que las imágenes de entrada, es decir, 784 neuronas de salida para generar una imagen de 28x28 píxeles en escala de grises."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         (None, 784)               0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 32)                25120     \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 784)               25872     \n",
      "=================================================================\n",
      "Total params: 50,992\n",
      "Trainable params: 50,992\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras.layers import Input, Dense\n",
    "from keras.models import Model\n",
    "\n",
    "# Tamanño de la representación interna del autoencoder\n",
    "encoding_dim = 32\n",
    "\n",
    "# capa de entrada\n",
    "input_img = Input(shape=(784,))\n",
    "# representación interna (encoded)\n",
    "encoded = Dense(encoding_dim, activation='relu')(input_img)\n",
    "# salida del autoencoder, es decir, la imagen reconstruida (decoded)\n",
    "decoded = Dense(784, activation='sigmoid')(encoded)\n",
    "\n",
    "# el modelo completo de autoencoder, que apila la entrada y la salida\n",
    "autoencoder = Model(input_img, decoded)\n",
    "\n",
    "autoencoder.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Para finalidades didácticas, definimos el modelo `encoder` que genera la codificación interna del autoencoder, sin el proceso de decodificación."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         (None, 784)               0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 32)                25120     \n",
      "=================================================================\n",
      "Total params: 25,120\n",
      "Trainable params: 25,120\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# este modelo crea la representación interna\n",
    "encoder = Model(input_img, encoded)\n",
    "\n",
    "encoder.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "De forma similar, definimos el modelo `decoder` que a partir de la representación interna (32 valores) reconstruye la imagen, es decir, aplica el proceso de decodificación."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_3 (InputLayer)         (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 784)               25872     \n",
      "=================================================================\n",
      "Total params: 25,872\n",
      "Trainable params: 25,872\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# capa de entrada de la representación interna\n",
    "encoded_input = Input(shape=(encoding_dim,))\n",
    "# obtener la última capa del autoencoder definido previamente\n",
    "decoder_layer = autoencoder.layers[-1]\n",
    "# creación del modelo\n",
    "decoder = Model(encoded_input, decoder_layer(encoded_input))\n",
    "\n",
    "decoder.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finalmente, definimos el optimizador y la función de pérdida empleada para el entrenamiento del *autoencoder*."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Entrenamiento del *autoencoder*\n",
    "\n",
    "En el siguiente fragmento de código se entrena el *autoencoder* creado anteriormente.\n",
    "\n",
    "Los hiperparámetros para ajustar el entrenamiento, además del optimizador y función de pérdida, son:\n",
    "- número de épocas del entrenamiento\n",
    "- tamaño del lote"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /usr/local/lib/python3.5/dist-packages/keras/backend/tensorflow_backend.py:1344: calling reduce_mean (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "keep_dims is deprecated, use keepdims instead\n"
     ]
    }
   ],
   "source": [
    "autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_epochs = 50\n",
    "n_batch_size = 256"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 60000 samples, validate on 10000 samples\n",
      "Epoch 1/50\n",
      "60000/60000 [==============================] - 3s 43us/step - loss: 0.3791 - val_loss: 0.2724\n",
      "Epoch 2/50\n",
      "60000/60000 [==============================] - 2s 38us/step - loss: 0.2649 - val_loss: 0.2543\n",
      "Epoch 3/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.2445 - val_loss: 0.2323\n",
      "Epoch 4/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.2243 - val_loss: 0.2141\n",
      "Epoch 5/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.2087 - val_loss: 0.2012\n",
      "Epoch 6/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1979 - val_loss: 0.1921\n",
      "Epoch 7/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1899 - val_loss: 0.1850\n",
      "Epoch 8/50\n",
      "60000/60000 [==============================] - 2s 38us/step - loss: 0.1833 - val_loss: 0.1789\n",
      "Epoch 9/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1774 - val_loss: 0.1733\n",
      "Epoch 10/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1722 - val_loss: 0.1684\n",
      "Epoch 11/50\n",
      "60000/60000 [==============================] - 2s 40us/step - loss: 0.1675 - val_loss: 0.1639\n",
      "Epoch 12/50\n",
      "60000/60000 [==============================] - 2s 38us/step - loss: 0.1633 - val_loss: 0.1599\n",
      "Epoch 13/50\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.1594 - val_loss: 0.1561\n",
      "Epoch 14/50\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.1558 - val_loss: 0.1527\n",
      "Epoch 15/50\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.1525 - val_loss: 0.1495\n",
      "Epoch 16/50\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.1493 - val_loss: 0.1464\n",
      "Epoch 17/50\n",
      "60000/60000 [==============================] - 3s 42us/step - loss: 0.1463 - val_loss: 0.1435\n",
      "Epoch 18/50\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.1435 - val_loss: 0.1407\n",
      "Epoch 19/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1409 - val_loss: 0.1382\n",
      "Epoch 20/50\n",
      "60000/60000 [==============================] - 2s 38us/step - loss: 0.1384 - val_loss: 0.1358\n",
      "Epoch 21/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1361 - val_loss: 0.1334\n",
      "Epoch 22/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1339 - val_loss: 0.1314\n",
      "Epoch 23/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1319 - val_loss: 0.1293\n",
      "Epoch 24/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1299 - val_loss: 0.1274\n",
      "Epoch 25/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1280 - val_loss: 0.1255\n",
      "Epoch 26/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1263 - val_loss: 0.1238\n",
      "Epoch 27/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1246 - val_loss: 0.1221\n",
      "Epoch 28/50\n",
      "60000/60000 [==============================] - 2s 40us/step - loss: 0.1229 - val_loss: 0.1206\n",
      "Epoch 29/50\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.1214 - val_loss: 0.1191\n",
      "Epoch 30/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1199 - val_loss: 0.1175\n",
      "Epoch 31/50\n",
      "60000/60000 [==============================] - 2s 38us/step - loss: 0.1185 - val_loss: 0.1162\n",
      "Epoch 32/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1172 - val_loss: 0.1149\n",
      "Epoch 33/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1159 - val_loss: 0.1137\n",
      "Epoch 34/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1148 - val_loss: 0.1126\n",
      "Epoch 35/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1137 - val_loss: 0.1116\n",
      "Epoch 36/50\n",
      "60000/60000 [==============================] - 2s 40us/step - loss: 0.1127 - val_loss: 0.1106\n",
      "Epoch 37/50\n",
      "60000/60000 [==============================] - 2s 40us/step - loss: 0.1117 - val_loss: 0.1097\n",
      "Epoch 38/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1109 - val_loss: 0.1088\n",
      "Epoch 39/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1101 - val_loss: 0.1080\n",
      "Epoch 40/50\n",
      "60000/60000 [==============================] - 2s 40us/step - loss: 0.1093 - val_loss: 0.1073\n",
      "Epoch 41/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1086 - val_loss: 0.1066\n",
      "Epoch 42/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1079 - val_loss: 0.1060\n",
      "Epoch 43/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1073 - val_loss: 0.1054\n",
      "Epoch 44/50\n",
      "60000/60000 [==============================] - 2s 40us/step - loss: 0.1068 - val_loss: 0.1049\n",
      "Epoch 45/50\n",
      "60000/60000 [==============================] - 2s 40us/step - loss: 0.1062 - val_loss: 0.1044\n",
      "Epoch 46/50\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.1057 - val_loss: 0.1038\n",
      "Epoch 47/50\n",
      "60000/60000 [==============================] - 2s 38us/step - loss: 0.1052 - val_loss: 0.1034\n",
      "Epoch 48/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1048 - val_loss: 0.1030\n",
      "Epoch 49/50\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.1044 - val_loss: 0.1026\n",
      "Epoch 50/50\n",
      "60000/60000 [==============================] - 2s 40us/step - loss: 0.1040 - val_loss: 0.1022\n"
     ]
    }
   ],
   "source": [
    "mfit = autoencoder.fit(x_train, x_train,\n",
    "                epochs=n_epochs,\n",
    "                batch_size=n_batch_size,\n",
    "                shuffle=True,\n",
    "                validation_data=(x_test, x_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY4AAAEaCAYAAAAG87ApAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nwSGEEKJBJDiEEEI0iASHEEKIBvn/TlmifySBwNIAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f957a6d8a90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# Plot del training loss\n",
    "plt.style.use(\"ggplot\")\n",
    "\n",
    "plt.plot(np.arange(0, n_epochs), mfit.history[\"loss\"], label=\"train\")\n",
    "plt.plot(np.arange(0, n_epochs), mfit.history[\"val_loss\"], label=\"val\")\n",
    "plt.title(\"Loss\")\n",
    "plt.xlabel(\"Epoch #\")\n",
    "plt.ylabel(\"Loss\")\n",
    "plt.legend(loc=\"upper right\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Visualización de los resultados\n",
    "\n",
    "Finalmente, presentaremos los resultados de aplicar el *autoencoder* en algunos ejemplos del conjunto de datos de test."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# codificamos y decodificamos las imágenes de test\n",
    "encoded_imgs = encoder.predict(x_test)\n",
    "decoded_imgs = decoder.predict(encoded_imgs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Notese que este paso se puede realizar en un solo paso, empleando:\n",
    "\n",
    "> decoded_imgs = autoencoder.predict(x_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABHEAAADqCAYAAAAlBtnSAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f958005fbe0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "n = 10  # cantidad de ejemplos que mostrar\n",
    "plt.figure(figsize=(20, 4))\n",
    "\n",
    "for i in range(n):\n",
    "    # mostrar la imagen 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",
    "    # mostrar la reconstrucción\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",
    "\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}