ANN/MLP_classification_keras_fashion_mnist.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import tensorflow as tf\n",
    "import keras # Don't import keras from tensorflow like (from tensorflow import keras)\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Classification using Multi-layer perceptron on Fashion-MNIST dataset.\n",
    "Fashion MNIST dataset holds 70,000 clothing images. Images value are stored in CSV files (train size: 60,000, test size: 10,000).  \n",
    "First column holds a number which is mapped to a particular cloth, other 784 columns hold corresponding pixel values.  \n",
    "0\tT-shirt/top  \n",
    "1\tTrouser  \n",
    "2\tPullover  \n",
    "3\tDress  \n",
    "4\tCoat  \n",
    "5\tSandal  \n",
    "6\tShirt  \n",
    "7\tSneaker  \n",
    "8\tBag  \n",
    "9\tAnkle boot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Reading train and test data\n",
    "train_data = pd.read_csv('fashion-mnist_train.csv')\n",
    "test_data = pd.read_csv('fashion-mnist_test.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Class names\n",
    "class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat', \n",
    "               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Making validation data out of test data\n",
    "val_data = test_data.iloc[:5000,:]\n",
    "test_data = test_data.iloc[5000:,:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Fetching the labels\n",
    "train_labels = train_data.label\n",
    "val_labels = val_data.label\n",
    "test_labels = test_data.label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(60000, 28, 28)"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Reshaping training data\n",
    "train_images = train_data.iloc[:,1:].values.reshape(60000, 28, 28)\n",
    "train_images.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(5000, 28, 28)"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Reshaping validation data\n",
    "val_images = val_data.iloc[:,1:].values.reshape(5000, 28, 28)\n",
    "val_images.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(5000, 28, 28)"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Reshaping test data\n",
    "test_images = test_data.iloc[:,1:].values.reshape(5000, 28, 28)\n",
    "test_images.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x223d89454a8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Confirming if given data is labelled correctly by plotting images against their label names\n",
    "plt.figure(figsize=(12,6))\n",
    "for i in range(1,7):\n",
    "    ax1 = plt.subplot(2,3,i)\n",
    "    ax1.imshow(train_images[i])\n",
    "    plt.xlabel(class_names[train_labels[i]])\n",
    "    plt.xticks([]) # removing xticks\n",
    "    plt.yticks([]) # removing yticks    \n",
    "    sns.despine(left=True, bottom=True) # removing borders\n",
    "# Inference - Given data is correct. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[255 254]\n",
      "[254 255]\n"
     ]
    }
   ],
   "source": [
    "# Finding pixel range values for all the images\n",
    "print(pd.unique([np.ptp(i) for i in train_images]))\n",
    "print(pd.unique([np.ptp(i) for i in test_images]))\n",
    "# Inference - Only values are 255 and 254 which means that color in all images fall in the range of 0-255."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Scaling data in the range of 0-1\n",
    "train_images = train_images/255.0\n",
    "val_images = val_images/255.0\n",
    "test_images = test_images/255.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Defining multi-layer perceptron model with 1 hidden layer having 256 neurons and output layer with 10 possible answers\n",
    "model = keras.Sequential([\n",
    "    keras.layers.Flatten(input_shape=(28, 28)), # Perform conversion of higher dimensional data (here, 2-D) to 1-D data.\n",
    "    keras.layers.Dense(256, activation=tf.nn.relu), # Hidden layer with 256 neurons and ReLU activation function\n",
    "    keras.layers.Dense(10, activation=tf.nn.softmax) # Output layer with softmax activation function \n",
    "])                                                   # which gives final output in terms of probability. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Prefer Average Cross Entropy over Classification error or MSE. Refer [here](https://jamesmccaffrey.wordpress.com/2013/11/05/why-you-should-use-cross-entropy-error-instead-of-classification-error-or-mean-squared-error-for-neural-network-classifier-training/) for more explanation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Defining parameters like optmizer, loss function and evaluating metric\n",
    "model.compile(loss='sparse_categorical_crossentropy',\n",
    "              optimizer=keras.optimizers.Adam(), # Learning rate and momentum can be passed inside optimizer\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 60000 samples, validate on 5000 samples\n",
      "Epoch 1/5\n",
      "60000/60000 [==============================] - 11s 182us/step - loss: 0.2192 - acc: 0.9178 - val_loss: 0.3097 - val_acc: 0.8850\n",
      "\n",
      "Epoch 00001: val_acc improved from -inf to 0.88500, saving model to MLP_checkpoint.hdf5\n",
      "Epoch 2/5\n",
      "60000/60000 [==============================] - 11s 182us/step - loss: 0.2156 - acc: 0.9188 - val_loss: 0.3002 - val_acc: 0.8940\n",
      "\n",
      "Epoch 00002: val_acc improved from 0.88500 to 0.89400, saving model to MLP_checkpoint.hdf5\n",
      "Epoch 3/5\n",
      "60000/60000 [==============================] - 12s 193us/step - loss: 0.2083 - acc: 0.9225 - val_loss: 0.3030 - val_acc: 0.8946\n",
      "\n",
      "Epoch 00003: val_acc improved from 0.89400 to 0.89460, saving model to MLP_checkpoint.hdf5\n",
      "Epoch 4/5\n",
      "60000/60000 [==============================] - 11s 186us/step - loss: 0.2005 - acc: 0.9248 - val_loss: 0.3121 - val_acc: 0.8916\n",
      "\n",
      "Epoch 00004: val_acc did not improve\n",
      "Epoch 5/5\n",
      "60000/60000 [==============================] - 12s 196us/step - loss: 0.1931 - acc: 0.9286 - val_loss: 0.3099 - val_acc: 0.8904\n",
      "\n",
      "Epoch 00005: val_acc did not improve\n"
     ]
    }
   ],
   "source": [
    "# Traning the model and writing log file for TensorBoard\n",
    "\n",
    "logdir = r'MLP'\n",
    "cpfile = r'MLP_checkpoint.hdf5' # Weights to be stored in HDF5 format\n",
    "\n",
    "# To store tensorboard graph logs\n",
    "cb_tensorboard = keras.callbacks.TensorBoard(log_dir=logdir, write_graph=True) \n",
    "\n",
    "# To store model checkpoint every time validation accuracy improves. You can assign different metrics to monitor like -\n",
    "# acc: Train accuracy\n",
    "# val_acc: Validation accuracy\n",
    "# loss: Train loss\n",
    "# val_loss: Validation loss\n",
    "cb_checkpoint = keras.callbacks.ModelCheckpoint(cpfile, monitor='val_acc', verbose=1, save_best_only=True, mode='max')\n",
    "\n",
    "epochs = 5\n",
    "history = model.fit(train_images, train_labels, epochs=epochs, \n",
    "                   validation_data=(val_images, val_labels),\n",
    "                   callbacks=[cb_tensorboard, cb_checkpoint])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Next time we can directly load the checkpoint model without waiting for model to train\n",
    "model.load_weights(\"MLP_checkpoint.hdf5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5000/5000 [==============================] - 0s 67us/step\n",
      "Test accuracy: 0.8992\n"
     ]
    }
   ],
   "source": [
    "# Evaluating the model on test data\n",
    "test_loss, test_acc = model.evaluate(test_images, test_labels)\n",
    "\n",
    "print('Test accuracy:', test_acc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Prediction\n",
    "predictions = model.predict(test_images)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x223a46c1240>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Original Label: Coat\n",
      "Predicted Label: Coat\n"
     ]
    }
   ],
   "source": [
    "# Visualizing the probability against target classes\n",
    "\n",
    "instance = 500 # choose an index out of the available 5000 test images \n",
    "out = pd.DataFrame(predictions[instance], index=class_names, columns=['Predicted Label'])\n",
    "out.plot.barh()\n",
    "plt.title('Original Label: '+class_names[test_labels.iloc[instance]])\n",
    "plt.show()\n",
    "\n",
    "print('Original Label: '+ class_names[test_labels.iloc[instance]])\n",
    "print('Predicted Label: ' + class_names[np.argmax(predictions[instance])]) # Predicted class with highest probability"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Save model to disk\n",
    "model.save('model.h5')\n",
    "\n",
    "# Load model from the disk\n",
    "loaded_model = keras.models.load_model('model.h5')"
   ]
  }
 ],
 "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}