mnist.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# A Very Simple Neural Network for MNIST Data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Import all necessary modules"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import sys\n",
    "from time import ctime, time\n",
    "if sys.version_info < (3,6):\n",
    "    raise Exception('This notebook runs with Python 3.6 and higher!')\n",
    "from time import time\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib as mpl\n",
    "import mnist\n",
    "\n",
    "import keras\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Activation,Lambda\n",
    "from keras.regularizers import l2,l1\n",
    "from keras.optimizers import Adam\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.initializers import zeros,glorot_uniform\n",
    "from keras.backend import one_hot, mean\n",
    "from keras.callbacks import BaseLogger, Callback, LearningRateScheduler\n",
    "\n",
    "import tensorflow as tf\n",
    "from IPython.display import clear_output\n",
    "\n",
    "%matplotlib inline\n",
    "plt.rcParams[\"figure.figsize\"] = (20,8)            \n",
    "tf.logging.set_verbosity(tf.logging.ERROR)\n",
    "np.random.seed(1)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def convert_to_one_hot(Y, C):\n",
    "    Y = np.eye(C)[Y.reshape(-1)]\n",
    "    return Y\n",
    "\n",
    "def imgshow(X, Y, index):\n",
    "    img = X[index].copy().reshape(28,28)\n",
    "    fig = plt.figure()\n",
    "    ax = fig.add_subplot(1,1,1)\n",
    "    imgplot = ax.imshow(img, cmap=mpl.cm.Greys)\n",
    "    imgplot.set_interpolation('nearest')\n",
    "    ax.xaxis.set_ticks_position('top')\n",
    "    ax.yaxis.set_ticks_position('left')\n",
    "    fig.text(.5, .08, f'Classified as {Y[index].nonzero()[0][0]}', \n",
    "             ha='center',\n",
    "            fontsize=15)\n",
    "    plt.show()\n",
    "    \n",
    "    \n",
    "class LearningRateCallBack(LearningRateScheduler):\n",
    "    def __init__(self,lrlist):\n",
    "        \"\"\"\n",
    "            learningrate : the learning rate for the model\n",
    "            lrlist : A np.arrays of shape (n,2) \n",
    "                    the 1st col is filled with epoch noumbers and the 2nd col is\n",
    "                    the associated lr\n",
    "            \n",
    "        \"\"\"\n",
    "        self.lrlist = lrlist\n",
    "        \n",
    "    def on_epoch_begin(self, epoch, logs=None):  \n",
    "        n = np.argmax(epoch>=self.lrlist.T[0])\n",
    "        return self.lrlist.T[1,n]\n",
    "    \n",
    "class LifePlots(keras.callbacks.Callback):\n",
    "    def __init__(self, epochs):\n",
    "        self.epochs = epochs\n",
    "    \n",
    "    def on_train_begin(self, logs={}):\n",
    "        self.time = time()\n",
    "        self.i = 0\n",
    "        self.x = []\n",
    "        self.losses = []\n",
    "        self.val_losses = []\n",
    "        self.logs = []\n",
    "        self.acc = []\n",
    "        self.val_acc = []\n",
    "\n",
    "    def on_epoch_end(self, epoch, logs={}):\n",
    "        \n",
    "        self.logs.append(logs)\n",
    "        self.x.append(self.i)\n",
    "        self.losses.append(logs.get('loss'))\n",
    "        self.val_losses.append(logs.get('val_loss'))\n",
    "        self.acc.append(logs.get('acc'))\n",
    "        self.val_acc.append(logs.get('val_acc'))\n",
    "        \n",
    "        \n",
    "        if (self.i%10 == 0) or (self.i==self.epochs-1) or (self.i==0):\n",
    "            clear_output(wait=True)\n",
    "            fig = plt.figure()\n",
    "            ax1 = fig.add_subplot(1,2,1)\n",
    "            ax1.set_title('Losses')\n",
    "            ax1.set_xlabel('Epoch')\n",
    "            ax1.set_xlim([0,self.epochs])\n",
    "            ax1.set_ylim([0.0,self.losses[0]])\n",
    "            ax1.plot(self.x, self.losses, label=\"loss\")\n",
    "            ax1.plot(self.x, self.val_losses, label=\"validation set loss\")\n",
    "            ax1.legend()\n",
    "\n",
    "            ax2 = fig.add_subplot(1,2,2)\n",
    "            ax2.set_title('Accuracy')\n",
    "            ax2.set_xlabel('Epoch')\n",
    "            ax2.set_xlim([0,self.epochs])\n",
    "            ax2.set_ylim([0,1.])\n",
    "            ax2.plot(self.x, self.acc, label=\"accuracy\")\n",
    "            ax2.plot(self.x, self.val_acc, label=\"validation set acc\")\n",
    "            ax2.legend()\n",
    "            plt.show();\n",
    "            xtime = time()-self.time\n",
    "            print(f\"Acc:    {self.acc[-1]:8.4f} Loss:    {self.losses[-1]:8.4f}\")            \n",
    "            print(f\"Dev-Acc:{self.val_acc[-1]:8.4f} Dev-Loss:{self.val_losses[-1]:8.4f}\")\n",
    "            print(f\"Seconds:{xtime:8.2f}\")\n",
    "            if self.i>=1:\n",
    "                print(f\"Estimate:{self.epochs/self.i*xtime:8.2f}\")\n",
    "                \n",
    "        \n",
    "        \n",
    "        self.i += 1\n",
    "\n",
    "callback_plot = None\n",
    "\n",
    "def init_plot(epochs):\n",
    "    global callback_plot\n",
    "    \n",
    "    callback_plot = LifePlots(epochs)\n",
    "    return callback_plot\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Read the data from files and reshape the data for the neural network"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "Y_train,X_train = mnist.read(dataset='training')\n",
    "Y_test,X_test = mnist.read(dataset='testing')\n",
    "\n",
    "Y_train = convert_to_one_hot(Y_train,10)\n",
    "Y_test = convert_to_one_hot(Y_test,10)\n",
    "\n",
    "# Data sizes\n",
    "N_train = X_train.shape[0]\n",
    "N_test = X_test.shape[0] // 2 # This is for the later split \n",
    "\n",
    "# Picture size\n",
    "pic_x = X_train.shape[1]\n",
    "pic_y = X_train.shape[2]\n",
    "N_pixels = pic_x * pic_y\n",
    "\n",
    "# Reshape the data\n",
    "X_train = X_train.reshape([N_train,N_pixels])\n",
    "X_test = X_test.reshape([X_test.shape[0],N_pixels])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The test set is consisting of two parts. The first half has centered numbers and the second half isn't preprocessed, which is basically a different distribution."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X train (60000, 784)\n",
      "Y train (60000, 10)\n",
      "X dev (5000, 784)\n",
      "Y dev (5000, 10)\n",
      "X test (5000, 784)\n",
      "Y test (5000, 10)\n"
     ]
    }
   ],
   "source": [
    "# Split the test into dev and test set\n",
    "X_dev = X_test[N_test:]\n",
    "Y_dev = Y_test[N_test:]\n",
    "X_test = X_test[0:N_test]\n",
    "Y_test = Y_test[0:N_test]\n",
    "\n",
    "print(f'X train {X_train.shape}')\n",
    "print(f'Y train {Y_train.shape}')\n",
    "\n",
    "print(f'X dev {X_dev.shape}')\n",
    "print(f'Y dev {Y_dev.shape}')\n",
    "\n",
    "print(f'X test {X_test.shape}')\n",
    "print(f'Y test {Y_test.shape}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Hyperparameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# These parameters are fitted for GPU. \n",
    "lamda  = 0.004\n",
    "lr     = 0.0004\n",
    "epochs = 1200\n",
    "batch  = 20000\n",
    "seed   = 234\n",
    "\n",
    "# Learningrates \n",
    "# It's an arrays with epoch, learning rate pairs\n",
    "lrarr = np.array(\n",
    "    [\n",
    "        [0,lr],\n",
    "        [600,lr/2],\n",
    "        [900,lr/4]\n",
    "    ]\n",
    "    )\n",
    "# This gives the number of nodes for all layers from 1 to N-1. The output layer is not included\n",
    "layerdata = [24,24] \n",
    "\n",
    "# Choose the activation function\n",
    "activation_func = 'relu'\n",
    "\n",
    "# Tensorboard logging\n",
    "logdata = True"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABI4AAAHwCAYAAADAeySPAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f89ac1391d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Acc:      0.9764 Loss:      0.2357\n",
      "Dev-Acc:  0.9770 Dev-Loss:  0.2239\n",
      "Seconds:  419.48\n",
      "Estimate:  419.83\n"
     ]
    }
   ],
   "source": [
    "model = Sequential()\n",
    "\n",
    "# Layer 1\n",
    "model.add(BatchNormalization(input_shape=(N_pixels,)))\n",
    "model.add(Dense(units=layerdata[0], \n",
    "                kernel_regularizer=l2(lamda),\n",
    "                #bias_regularizer=l2(lamda),                \n",
    "                kernel_initializer=glorot_uniform(seed=seed),\n",
    "                bias_initializer=zeros()))\n",
    "\n",
    "model.add(Activation(activation_func))\n",
    "\n",
    "# Layers from 2 to N-1\n",
    "for u in layerdata[1:]:\n",
    "    model.add(Dense(units=u, \n",
    "                    kernel_regularizer=l2(lamda),\n",
    "                    #bias_regularizer=l2(lamda),                \n",
    "                    kernel_initializer=glorot_uniform(seed=seed),\n",
    "                    bias_initializer=zeros()))\n",
    "\n",
    "    model.add(Activation(activation_func))\n",
    "    \n",
    "# Output layer\n",
    "model.add(Dense(\n",
    "        units=10,\n",
    "        kernel_regularizer=l2(lamda),\n",
    "        #bias_regularizer=l2(lamda),                \n",
    "        kernel_initializer=glorot_uniform(seed=seed),\n",
    "        bias_initializer=zeros()\n",
    "        ))\n",
    "\n",
    "model.add(Activation('softmax'))\n",
    "\n",
    "\n",
    "# Train the model\n",
    "model.compile(optimizer=Adam(lr=lr, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0),\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "init_plot(epochs)\n",
    "\n",
    "callbacks = [callback_plot, LearningRateCallBack(lrarr)]\n",
    "\n",
    "if logdata:\n",
    "    tbCallBack = keras.callbacks.TensorBoard(\n",
    "        log_dir=f'./log/{ctime()}', \n",
    "        batch_size=batch,\n",
    "        histogram_freq=1, \n",
    "        write_graph=True, \n",
    "        write_images=False)\n",
    "    callbacks.append(tbCallBack)\n",
    "\n",
    "\n",
    "history = model.fit(X_train, Y_train, \n",
    "                    validation_data=[X_dev,Y_dev],\n",
    "                    epochs=epochs, \n",
    "                    batch_size=batch, \n",
    "                    callbacks=callbacks,\n",
    "                    verbose=0, \n",
    "                    shuffle=True)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluate the test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The accuracy for the test set is    0.953\n"
     ]
    }
   ],
   "source": [
    "result = model.evaluate(X_test, Y_test, batch_size=batch, verbose=0)\n",
    "print(f\"The accuracy for the test set is {result[1]:8.3f}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Check an image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAdgAAAHvCAYAAAD3ps6MAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f89cf5e72e8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "index = 123\n",
    "\n",
    "imgshow(X_test,Y_test, index)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Get the prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The prediction for this image is 6\n"
     ]
    }
   ],
   "source": [
    "prediction = np.argmax(model.predict(X_test[index].reshape([1,N_pixels])))\n",
    "print(f'The prediction for this image is {prediction}')"
   ]
  }
 ],
 "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}