Autoencoder_1.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "02d0a567",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.layers import Dense, Input\n",
    "from tensorflow.keras.layers import Conv2D, Flatten\n",
    "from tensorflow.keras.layers import Reshape, Conv2DTranspose\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.datasets import mnist \n",
    "from tensorflow.keras.utils import plot_model \n",
    "from tensorflow.keras import backend as K "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "e8897d3a",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "9fe9f250",
   "metadata": {},
   "outputs": [],
   "source": [
    "#loading dataset MNIST \n",
    "\n",
    "(x_train, _), (x_test,_) = mnist.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "bde42c07",
   "metadata": {},
   "outputs": [],
   "source": [
    "#reshape to (28,28,1) and normalize input images \n",
    "\n",
    "image_size = x_train.shape[1]\n",
    "x_train = np.reshape(x_train, [-1, image_size, image_size,1])\n",
    "x_test = np.reshape(x_test, [-1, image_size, image_size, 1])\n",
    "\n",
    "x_train = x_train.astype('float32') /255 \n",
    "x_test = x_test.astype('float32') / 255\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "0d8fbaf7",
   "metadata": {},
   "outputs": [],
   "source": [
    "#network parameters \n",
    "\n",
    "input_shape = (image_size, image_size, 1)\n",
    "batch_size = 32 \n",
    "kernel_size = 3\n",
    "latent_dim = 16 \n",
    "\n",
    "#encoder/decoder number of CNN layers and filters per layer\n",
    "layer_filters = [32, 64]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "5aac4c22",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Building the autoencode MODEL \n",
    "\n",
    "#First - the Encoder part \n",
    "\n",
    "inputs = Input(shape=input_shape, name='encoder_input')\n",
    "x = inputs \n",
    "\n",
    "#stack of Conv2D(32) - Conv2d(64)\n",
    "for filters in layer_filters:\n",
    "    \n",
    "    x = Conv2D(filters = filters,\n",
    "              kernel_size=kernel_size,\n",
    "              activation='relu',\n",
    "              strides = 2,\n",
    "              padding='same')(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "132b6ce9",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Shape info needed to build decoder model so we dont \n",
    "#hand computation the input to the decoder's first\n",
    "# Conv2DTranspose will have this shpae\n",
    "#shape is (7,7,64) which is processed by the decoder\n",
    "#back to (28,28,1)\n",
    "\n",
    "shape = K.int_shape(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "ef003425",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"encoder\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " encoder_input (InputLayer)  [(None, 28, 28, 1)]       0         \n",
      "                                                                 \n",
      " conv2d (Conv2D)             (None, 14, 14, 32)        320       \n",
      "                                                                 \n",
      " conv2d_1 (Conv2D)           (None, 7, 7, 64)          18496     \n",
      "                                                                 \n",
      " flatten (Flatten)           (None, 3136)              0         \n",
      "                                                                 \n",
      " latent_vector (Dense)       (None, 16)                50192     \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 69,008\n",
      "Trainable params: 69,008\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "#generate latent vector \n",
    "\n",
    "x = Flatten()(x)\n",
    "latent = Dense(latent_dim, name='latent_vector')(x)\n",
    "\n",
    "#instantiate encoder model \n",
    "\n",
    "encoder = Model(inputs, latent, name='encoder')\n",
    "encoder.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "68bf07a1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "You must install pydot (`pip install pydot`) and install graphviz (see instructions at https://graphviz.gitlab.io/download/) for plot_model/model_to_dot to work.\n"
     ]
    }
   ],
   "source": [
    "plot_model(encoder, to_file='encoder.png', show_shapes=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "3a82a335",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Requirement already satisfied: pydot in c:\\python39\\lib\\site-packages (1.4.2)\n",
      "Requirement already satisfied: pyparsing>=2.1.4 in c:\\python39\\lib\\site-packages (from pydot) (3.0.8)\n"
     ]
    }
   ],
   "source": [
    "!pip install pydot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "692c0a57",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Requirement already satisfied: graphviz in c:\\python39\\lib\\site-packages (0.20.1)\n"
     ]
    }
   ],
   "source": [
    "!pip install graphviz"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "85bb01c0",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "You must install pydot (`pip install pydot`) and install graphviz (see instructions at https://graphviz.gitlab.io/download/) for plot_model/model_to_dot to work.\n"
     ]
    }
   ],
   "source": [
    "plot_model(encoder, to_file='encoder.png', show_shapes=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "9db81018",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Build the decoder model\n",
    "\n",
    "latent_inputs = Input(shape=(latent_dim,),name='decoder_input')\n",
    "\n",
    "#use the shape (7,7,64) that was earlier saved\n",
    "x = Dense(shape[1]*shape[2]*shape[3])(latent_inputs)\n",
    "\n",
    "#from vector to suitable shape for tansposed conv\n",
    "x = Reshape((shape[1], shape[2],shape[3]))(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "24c26e17",
   "metadata": {},
   "outputs": [],
   "source": [
    "#stack of Conv2DTranspose(64)-conv2DTranpose(32)\n",
    "\n",
    "for filters in layer_filters[::-1]:\n",
    "    \n",
    "    x = Conv2DTranspose(filters=filters,\n",
    "                       kernel_size=kernel_size,\n",
    "                       activation='relu',\n",
    "                       strides=2,\n",
    "                       padding='same')(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "0686ae00",
   "metadata": {},
   "outputs": [],
   "source": [
    "#Reconstruct the input\n",
    "\n",
    "outputs = Conv2DTranspose(filters=1,\n",
    "                         kernel_size=kernel_size,\n",
    "                         activation='sigmoid',\n",
    "                         padding='same',\n",
    "                         name='decoder_output')(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "68ec731c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"decoder\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " decoder_input (InputLayer)  [(None, 16)]              0         \n",
      "                                                                 \n",
      " dense (Dense)               (None, 3136)              53312     \n",
      "                                                                 \n",
      " reshape (Reshape)           (None, 7, 7, 64)          0         \n",
      "                                                                 \n",
      " conv2d_transpose (Conv2DTra  (None, 14, 14, 64)       36928     \n",
      " nspose)                                                         \n",
      "                                                                 \n",
      " conv2d_transpose_1 (Conv2DT  (None, 28, 28, 32)       18464     \n",
      " ranspose)                                                       \n",
      "                                                                 \n",
      " decoder_output (Conv2DTrans  (None, 28, 28, 1)        289       \n",
      " pose)                                                           \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 108,993\n",
      "Trainable params: 108,993\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "#instantiate decoder model \n",
    "decoder = Model(latent_inputs,outputs, name='decoder')\n",
    "decoder.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "68f2659a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"autoencoder\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " encoder_input (InputLayer)  [(None, 28, 28, 1)]       0         \n",
      "                                                                 \n",
      " encoder (Functional)        (None, 16)                69008     \n",
      "                                                                 \n",
      " decoder (Functional)        (None, 28, 28, 1)         108993    \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 178,001\n",
      "Trainable params: 178,001\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "#autoencoder = encoder + decoder\n",
    "#instantiate autoencoder model \n",
    "\n",
    "autoencoder = Model(inputs,\n",
    "                   decoder(encoder(inputs)),\n",
    "                   name='autoencoder')\n",
    "autoencoder.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "79c18f76",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "You must install pydot (`pip install pydot`) and install graphviz (see instructions at https://graphviz.gitlab.io/download/) for plot_model/model_to_dot to work.\n"
     ]
    }
   ],
   "source": [
    "plot_model(autoencoder, to_file='autoencoder.png',\n",
    "          show_shapes=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "74a29768",
   "metadata": {},
   "outputs": [],
   "source": [
    "#mean square error loss function and Adam optimizer \n",
    "\n",
    "autoencoder.compile(loss='mse',optimizer='adam')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "d7552dca",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/20\n",
      "1875/1875 [==============================] - 34s 18ms/step - loss: 0.0206 - val_loss: 0.0105\n",
      "Epoch 2/20\n",
      "1875/1875 [==============================] - 37s 20ms/step - loss: 0.0098 - val_loss: 0.0090\n",
      "Epoch 3/20\n",
      "1875/1875 [==============================] - 40s 22ms/step - loss: 0.0087 - val_loss: 0.0082\n",
      "Epoch 4/20\n",
      "1875/1875 [==============================] - 46s 24ms/step - loss: 0.0081 - val_loss: 0.0079\n",
      "Epoch 5/20\n",
      "1875/1875 [==============================] - 55s 29ms/step - loss: 0.0077 - val_loss: 0.0075\n",
      "Epoch 6/20\n",
      "1875/1875 [==============================] - 67s 36ms/step - loss: 0.0074 - val_loss: 0.0075\n",
      "Epoch 7/20\n",
      "1875/1875 [==============================] - 57s 30ms/step - loss: 0.0072 - val_loss: 0.0073\n",
      "Epoch 8/20\n",
      "1875/1875 [==============================] - 56s 30ms/step - loss: 0.0070 - val_loss: 0.0072\n",
      "Epoch 9/20\n",
      "1875/1875 [==============================] - 62s 33ms/step - loss: 0.0069 - val_loss: 0.0071\n",
      "Epoch 10/20\n",
      "1875/1875 [==============================] - 58s 31ms/step - loss: 0.0068 - val_loss: 0.0070\n",
      "Epoch 11/20\n",
      "1875/1875 [==============================] - 60s 32ms/step - loss: 0.0067 - val_loss: 0.0069\n",
      "Epoch 12/20\n",
      "1875/1875 [==============================] - 58s 31ms/step - loss: 0.0066 - val_loss: 0.0068\n",
      "Epoch 13/20\n",
      "1875/1875 [==============================] - 60s 32ms/step - loss: 0.0065 - val_loss: 0.0068\n",
      "Epoch 14/20\n",
      "1875/1875 [==============================] - 58s 31ms/step - loss: 0.0065 - val_loss: 0.0067\n",
      "Epoch 15/20\n",
      "1875/1875 [==============================] - 60s 32ms/step - loss: 0.0064 - val_loss: 0.0067\n",
      "Epoch 16/20\n",
      "1875/1875 [==============================] - 59s 31ms/step - loss: 0.0063 - val_loss: 0.0067\n",
      "Epoch 17/20\n",
      "1875/1875 [==============================] - 60s 32ms/step - loss: 0.0063 - val_loss: 0.0066\n",
      "Epoch 18/20\n",
      "1875/1875 [==============================] - 58s 31ms/step - loss: 0.0062 - val_loss: 0.0065\n",
      "Epoch 19/20\n",
      "1875/1875 [==============================] - 60s 32ms/step - loss: 0.0062 - val_loss: 0.0066\n",
      "Epoch 20/20\n",
      "1875/1875 [==============================] - 58s 31ms/step - loss: 0.0062 - val_loss: 0.0065\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x239c73eefa0>"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#train the autoencoder\n",
    "autoencoder.fit(x_train, x_train,\n",
    "               validation_data=(x_test,x_test),\n",
    "               epochs=20,\n",
    "               batch_size=batch_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "22fca8c0",
   "metadata": {},
   "outputs": [],
   "source": [
    "#predict the autoencoder output from test data\n",
    "\n",
    "x_decoded = autoencoder.predict(x_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "ec2c1573",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "#display the first 10 input and decoded images\n",
    "\n",
    "imgs = np.concatenate([x_test[:8], x_decoded[:8]])\n",
    "imgs = imgs.reshape((4,4,image_size, image_size))\n",
    "imgs = np.vstack([np.hstack(i) for i in imgs])\n",
    "plt.figure()\n",
    "plt.axis('off')\n",
    "plt.title(\"Input first 2 rows. Decoded: last 2 rows\")\n",
    "plt.imshow(imgs, interpolation='none',cmap='gray')\n",
    "plt.savefig('input_n_decoded.png')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "681f360b",
   "metadata": {},
   "source": [
    "#### "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}