ConvolutionalNeuralNetwork.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "# Implementing Convolutional Neural Network from Scratch.\n",
    "**Data-Set Digit Recognizer from Kaggle = https://www.kaggle.com/c/digit-recognizer/data ** <br>\n",
    "**Digits 0-9**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Import the Required Libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd              #for data manipulation\n",
    "import numpy as np               #for maths\n",
    "import matplotlib.pyplot as plt  #for visualization\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Get the Training Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "#path to data\n",
    "path = r'E:\\Study n Work\\Machine Learning\\Kaggle\\Digit Recognizer\\Data\\train.csv'\n",
    "\n",
    "#read data\n",
    "train_data = pd.read_csv(path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>label</th>\n",
       "      <th>pixel0</th>\n",
       "      <th>pixel1</th>\n",
       "      <th>pixel2</th>\n",
       "      <th>pixel3</th>\n",
       "      <th>pixel4</th>\n",
       "      <th>pixel5</th>\n",
       "      <th>pixel6</th>\n",
       "      <th>pixel7</th>\n",
       "      <th>pixel8</th>\n",
       "      <th>...</th>\n",
       "      <th>pixel774</th>\n",
       "      <th>pixel775</th>\n",
       "      <th>pixel776</th>\n",
       "      <th>pixel777</th>\n",
       "      <th>pixel778</th>\n",
       "      <th>pixel779</th>\n",
       "      <th>pixel780</th>\n",
       "      <th>pixel781</th>\n",
       "      <th>pixel782</th>\n",
       "      <th>pixel783</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>3 rows × 785 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "   label  pixel0  pixel1  pixel2  pixel3  pixel4  pixel5  pixel6  pixel7  \\\n",
       "0      1       0       0       0       0       0       0       0       0   \n",
       "1      0       0       0       0       0       0       0       0       0   \n",
       "2      1       0       0       0       0       0       0       0       0   \n",
       "\n",
       "   pixel8    ...     pixel774  pixel775  pixel776  pixel777  pixel778  \\\n",
       "0       0    ...            0         0         0         0         0   \n",
       "1       0    ...            0         0         0         0         0   \n",
       "2       0    ...            0         0         0         0         0   \n",
       "\n",
       "   pixel779  pixel780  pixel781  pixel782  pixel783  \n",
       "0         0         0         0         0         0  \n",
       "1         0         0         0         0         0  \n",
       "2         0         0         0         0         0  \n",
       "\n",
       "[3 rows x 785 columns]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#lets see the data\n",
    "train_data.head(3)       "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of Dataset (42000, 785)\n",
      "Number of Images 42000\n",
      "Number of pixels 784\n",
      "Unique Labels    [0 1 2 3 4 5 6 7 8 9]\n"
     ]
    }
   ],
   "source": [
    "shape = train_data.shape  #get shape\n",
    "m = shape[0]              #number of images \n",
    "num_pix = shape[1]-1      #number of pixels, First Col is Label of Image so '-1'\n",
    "labels = np.unique(train_data['label'])\n",
    "\n",
    "print(\"Shape of Dataset {}\".format(shape))\n",
    "print(\"Number of Images {}\".format(m))\n",
    "print(\"Number of pixels {}\".format(num_pix))\n",
    "print(\"Unique Labels    {}\".format(labels))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Separating Features and Labels from the dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Shape of Labels   = (42000, 1)\n",
      "Shape of Features = (42000, 784)\n"
     ]
    }
   ],
   "source": [
    "#Labels\n",
    "train_labels = np.array(train_data['label']).reshape(train_data.shape[0],1)\n",
    "print(\"Shape of Labels   = {}\".format(train_labels.shape))\n",
    "\n",
    "#Features\n",
    "train_dataset = np.array(train_data.iloc[:,1:785]).reshape(train_data.shape[0],784)\n",
    "print(\"Shape of Features = {}\".format(train_dataset.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Lets see some images"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Labels :  [7]   [5]   [1]   [9]   [0]   "
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAABcCAYAAAB+6068AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ntlXV0e2BegAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x2078052ec18>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "images = np.array(train_dataset[50:55,:]).reshape(5,28,28,1)   #convert it into 28*28*1\n",
    "labs = np.array(train_labels[50:55]).reshape(5,1)\n",
    "\n",
    "fig = plt.figure()\n",
    "print(\"Labels : \",end= ' ')\n",
    "for i in range(5):\n",
    "    fig.add_subplot(1,5,i+1)\n",
    "    plt.imshow(images[i,:,:,0],cmap='gray')\n",
    "    print(\"{}  \".format(labs[i]),end=\" \")\n",
    "    plt.axis(\"off\")\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Normalize the dataset and Break the dataset into Test, Valid, Train dataset\n",
    "\n",
    "**Normalize :**\n",
    "image_data = (image_data-127)/255\n",
    "\n",
    "**Data Distribution**\n",
    "1. Train Dataset - 70%\n",
    "2. Valid Dataset - 15%\n",
    "3. Test  Dataset - 15%\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#Normalizing the dataset\n",
    "train_dataset = (train_dataset-127)/255"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train dataset = (30000, 784)\n",
      "Valid dataset = (6000, 784)\n",
      "Test dataset  = (6000, 784)\n"
     ]
    }
   ],
   "source": [
    "#Breaking the dataset into test dataset, valid dataset and train dataset.\n",
    "\n",
    "test_dataset = np.array(train_dataset[36000:42000]).reshape(-1,num_pix)\n",
    "test_labels = np.array(train_labels[36000:42000]).reshape(-1,1)\n",
    "\n",
    "valid_dataset = np.array(train_dataset[30000:36000]).reshape(-1,num_pix)\n",
    "valid_labels = np.array(train_labels[30000:36000]).reshape(-1,1)\n",
    "\n",
    "train_dataset = np.array(train_dataset[0:30000]).reshape(-1,num_pix)\n",
    "train_labels = np.array(train_labels[0:30000]).reshape(-1,1)\n",
    "\n",
    "print(\"Train dataset = {}\".format(train_dataset.shape,train_labels.shape))\n",
    "print(\"Valid dataset = {}\".format(valid_dataset.shape,valid_labels.shape))\n",
    "print(\"Test dataset  = {}\".format(test_dataset.shape,test_labels.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### One Hot Encoding\n",
    "\n",
    "**Example : If label = 4 : [0,0,0,0,1,0,0,0,0,0]**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train labels = (30000, 10)\n",
      "Valid labels = (6000, 10)\n",
      "Test labels  = (6000, 10)\n"
     ]
    }
   ],
   "source": [
    "neural_train_labels = np.zeros((train_labels.shape[0],10))\n",
    "neural_valid_labels = np.zeros((valid_labels.shape[0],10))\n",
    "neural_test_labels = np.zeros((test_labels.shape[0],10))\n",
    "\n",
    "for i,value in enumerate(train_labels):\n",
    "    neural_train_labels[i,value] = 1\n",
    "\n",
    "for i,value in enumerate(valid_labels):\n",
    "    neural_valid_labels[i,value] = 1\n",
    "\n",
    "for i,value in enumerate(test_labels):\n",
    "    neural_test_labels[i,value] = 1\n",
    "\n",
    "print(\"Train labels = {}\".format(neural_train_labels.shape))\n",
    "print(\"Valid labels = {}\".format(neural_valid_labels.shape))\n",
    "print(\"Test labels  = {}\".format(neural_test_labels.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Reformatting the dataset\n",
    "**Example : pixels = 784 , convert them into [28,28,1]**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set   = (30000, 28, 28, 1)\n",
      "Validation set = (6000, 28, 28, 1)\n",
      "Test set       = (6000, 28, 28, 1)\n"
     ]
    }
   ],
   "source": [
    "image_size = 28\n",
    "num_labels = 10\n",
    "num_channels = 1 # grayscale\n",
    "\n",
    "def reformat(dataset):\n",
    "    dataset = dataset.reshape(-1,image_size,image_size,num_channels).astype(np.float32)\n",
    "    return dataset\n",
    "\n",
    "train_dataset = reformat(train_dataset)\n",
    "valid_dataset = reformat(valid_dataset)\n",
    "test_dataset = reformat(test_dataset)\n",
    "\n",
    "print('Training set   = {}'.format(train_dataset.shape, neural_train_labels.shape))\n",
    "print('Validation set = {}'.format(valid_dataset.shape, neural_valid_labels.shape))\n",
    "print('Test set       = {}'.format(test_dataset.shape, neural_test_labels.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Hyperparameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#pixels\n",
    "image_size = 28\n",
    "\n",
    "#greyscale\n",
    "num_channels = 1\n",
    "\n",
    "#patch size\n",
    "patch_size = 3\n",
    "\n",
    "#depth\n",
    "depth = 4\n",
    "\n",
    "#hidden layers\n",
    "hidden1 = 256\n",
    "\n",
    "#hyperparameters\n",
    "learning_rate = 0.1\n",
    "\n",
    "#regularization\n",
    "beta = 0\n",
    "\n",
    "#target_labels\n",
    "num_classes = 10"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Activation Functions\n",
    "**In-Between Layers** <br>\n",
    "* **Sigmoid = 1 / (1+exp(-X))** \n",
    "\n",
    "**For Final Layer**\n",
    "* **Softmax = exp(X) / (sum(exp(X),1))**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#sigmoid\n",
    "def sigmoid(X):\n",
    "    return 1/(1+np.exp(-1*X))\n",
    "\n",
    "#softmax\n",
    "def softmax(X):\n",
    "    exp_X = np.exp(X)\n",
    "    sum_exp_X = np.sum(exp_X,1).reshape(-1,1)  #col-wise sum\n",
    "    exp_X = exp_X/sum_exp_X\n",
    "    return exp_X"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Initialize Weights\n",
    "* ** conv_layer1_weights = [3,3,1,4] **<br>\n",
    "* ** conv_layer2_weights = [3,3,4,16]**<br>\n",
    "* ** full_layer1_weights = [504,256] **<br>\n",
    "* ** full_layer2_weights = [256,10]  **<br>\n",
    "\n",
    "Store these weights in parameters dictionary!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    " def initialize_parameters():\n",
    "    #initialize weights values with 0 mean and 0.5 standard deviation.\n",
    "    mean = 0\n",
    "    std = 0.5\n",
    "    \n",
    "    #conv layer weights\n",
    "    conv_layer1_weights = np.random.normal(mean,std,(patch_size,patch_size,num_channels,depth))\n",
    "    conv_layer1_biases = np.zeros([1,depth])\n",
    "    conv_layer2_weights = np.random.normal(mean,std,(patch_size,patch_size,depth,depth*4))\n",
    "    conv_layer2_biases = np.zeros([1,depth*4])\n",
    "    \n",
    "    #fully-connected weights\n",
    "    full_layer1_weights = np.random.normal(mean,std,(((image_size//4-1) * (image_size//4-1) * depth * 4),hidden1))\n",
    "    full_layer1_biases = np.zeros([hidden1])\n",
    "    full_layer2_weights = np.random.normal(mean,std,(hidden1,num_classes))\n",
    "    full_layer2_biases = np.zeros([num_classes])\n",
    "    \n",
    "    parameters = dict()\n",
    "    parameters['cw1'] = conv_layer1_weights\n",
    "    parameters['cb1'] = conv_layer1_biases\n",
    "    parameters['cw2'] = conv_layer2_weights\n",
    "    parameters['cb2'] = conv_layer2_biases\n",
    "    parameters['fw1'] = full_layer1_weights\n",
    "    parameters['fb1'] = full_layer1_biases\n",
    "    parameters['fw2'] = full_layer2_weights\n",
    "    parameters['fb2'] = full_layer2_biases\n",
    "    \n",
    "    return parameters"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Convolution Operation\n",
    "![](https://github.com/Ahmadkhan02/Neural-Network-from-scratch/blob/main/images/Convolution.PNG?raw=true>)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![]( \"https://github.com/navjindervirdee/neural-networks/blob/master/Convolutional%20Neural%20Network/Convolution.PNG?raw=true\" )\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* **Red   Cell : 5x1 + 4x2 + 3x1 + 6x3 + 2x1 + 1x0 + 5x2 + 7x3 + 3x1 = 70**\n",
    "* **Green Cell : 4x1 + 3x2 + 1x1 + 2x3 + 1x1 + 4x0 + 7x2 + 3x3 + 2x1 = 43**\n",
    "* **Blue  Cell : 6x1 + 2x2 + 1x1 + 5x3 + 7x1 + 3x0 + 1x2 + 4x3 + 2x1 = 49**\n",
    "* **Purp  Cell : 2x1 + 1x2 + 4x1 + 7x3 + 3x1 + 2x0 + 4x2 + 2x3 + 3x1 = 49**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the function below Hardcoded : <br>\n",
    "Stride = 2 <br>\n",
    "Padding = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#Convolution operation i.e multiplying Image with the weights.\n",
    "#stride hardcoded = 2\n",
    "#padding  = 0\n",
    "def conv_multiply(image,weights):\n",
    "    hsize = (image.shape[0]-weights.shape[0])//2 + 1\n",
    "    vsize = (image.shape[1]-weights.shape[1])//2 + 1\n",
    "    logits = np.zeros([hsize,vsize,weights.shape[3]])\n",
    "    for d in range(weights.shape[3]):\n",
    "        row = 0\n",
    "        for rpos in range(0,(image.shape[0]-patch_size+1),2):\n",
    "            col=0\n",
    "            for cpos in range(0,(image.shape[1]-patch_size+1),2):\n",
    "                logits[row,col,d] = np.sum(np.multiply(image[rpos:rpos+patch_size, cpos:cpos+patch_size, :],weights[:,:,:,d]))\n",
    "                col += 1\n",
    "            row+=1\n",
    "    return logits"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Forward Propagation\n",
    "![](https://github.com/Ahmadkhan02/Neural-Network-from-scratch/blob/main/images/Forward.JPG?raw=true)\n",
    "\n",
    "* Function returns the intermediate activations in **cache dictionary** and final output in output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#FORWARD PROPAGATION\n",
    "def forward_propagation(dataset,parameters):\n",
    "    #convolution layers activations\n",
    "    m = dataset.shape[0]\n",
    "    \n",
    "    #get the parameters\n",
    "    cw1 = parameters['cw1']\n",
    "    cb1 = parameters['cb1']\n",
    "    cw2 = parameters['cw2']\n",
    "    cb2 = parameters['cb2']\n",
    "    \n",
    "    fw1 = parameters['fw1']\n",
    "    fb1 = parameters['fb1']\n",
    "    fw2 = parameters['fw2']\n",
    "    fb2 = parameters['fb2']\n",
    "    \n",
    "    #to store the intermediate activations for backward propagation\n",
    "    cache = dict()\n",
    "    \n",
    "    conv_activation1 = list()\n",
    "    conv_activation2 = list()\n",
    "    \n",
    "    #image by image convolutional forward propagation\n",
    "    for i in range(m):\n",
    "        image = dataset[i]\n",
    "        logits = conv_multiply(image,cw1) + cb1\n",
    "        ca1 = sigmoid(logits)\n",
    "        ca2 = sigmoid(conv_multiply(ca1,cw2) + cb2).reshape((image_size // 4 -1) * (image_size // 4 -1) * depth * 4)\n",
    "        \n",
    "        conv_activation1.append(ca1)\n",
    "        conv_activation2.append(ca2)\n",
    "        \n",
    "    #convert into numpy array\n",
    "    conv_activation1 = np.array(conv_activation1).reshape(m,image_size // 2 -1, image_size // 2 -1, depth)\n",
    "    conv_activation2 = np.array(conv_activation2).reshape(m,image_size // 4 -1, image_size // 4 -1, depth * 4)\n",
    "        \n",
    "    #expand the conv_activation2 into (m,num_features) \n",
    "    #num_features = (image_size // 4 * image_size // 4 * depth * 4)\n",
    "    temp_activation = np.array(conv_activation2).reshape(m,(image_size // 4 -1) * (image_size // 4-1) * depth * 4)\n",
    "    \n",
    "    #fully connected layers activations\n",
    "    full_activation1 = np.matmul(temp_activation,fw1) + fb1\n",
    "    full_activation1 = sigmoid(full_activation1)\n",
    "    full_activation2 = np.matmul(full_activation1,fw2) + fb2\n",
    "    output = softmax(full_activation2)\n",
    "    \n",
    "    cache['ca1'] = conv_activation1\n",
    "    cache['ca2'] = conv_activation2\n",
    "    cache['fa1'] = full_activation1\n",
    "    cache['output'] = output\n",
    "    return cache,output"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Backward Propagation\n",
    "* Apply the Chain Rule and calculate the derivatives.\n",
    "* Returns the derivatives in ** derivatives dictionary **"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Calculate Error terms for Conv Layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#calculate conv deltas or errors only for one example\n",
    "def conv_delta(next_error,weights):\n",
    "    delta = np.zeros([next_error.shape[0]*2+1,next_error.shape[1]*2+1,next_error.shape[2]//4])\n",
    "    for d in range(weights.shape[3]):\n",
    "        row = 0\n",
    "        for rpos in range(0,delta.shape[0]-patch_size+1,2):\n",
    "            col=0\n",
    "            for cpos in range(0,delta.shape[2]-patch_size+1,2):\n",
    "                delta[rpos:rpos+patch_size,cpos:cpos+patch_size,:] += weights[:,:,:,d]*next_error[row,col,d]\n",
    "                col+=1\n",
    "            row +=1\n",
    "    return delta"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Calculate the derivatives for Conv layer Weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#conv partial derivatives only for single example\n",
    "def conv_derivatives(delta,activation):\n",
    "    partial_derivatives = np.zeros([patch_size,patch_size,activation.shape[2],delta.shape[2]])\n",
    "    for d2 in range(0,partial_derivatives.shape[3]):\n",
    "        row=0\n",
    "        for rpos in range(0,activation.shape[0]-patch_size+1,2):\n",
    "            col = 0\n",
    "            for cpos in range(0,activation.shape[1]-patch_size+1,2):\n",
    "                partial_derivatives[:,:,:,d2] += np.multiply(activation[rpos:rpos+patch_size, cpos:cpos+patch_size, :],delta[row,col,d2])\n",
    "                col += 1\n",
    "            row += 1\n",
    "    return partial_derivatives"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Full Backward Propagation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "def backward_propagation(dataset,labels,cache,parameters):\n",
    "    #get activations\n",
    "    output = cache['output']\n",
    "    fa1 = cache['fa1']\n",
    "    ca2 = cache['ca2']\n",
    "    ca1 = cache['ca1']\n",
    "    \n",
    "    temp_act = np.array(ca2).reshape(-1,(image_size // 4-1) * (image_size // 4 -1)* depth * 4)\n",
    "    \n",
    "    #get parameters\n",
    "    cw1 = parameters['cw1']\n",
    "    cw2 = parameters['cw2']\n",
    "    fw1 = parameters['fw1']\n",
    "    fw2 = parameters['fw2']\n",
    "    \n",
    "    \n",
    "    #cal errors fully connected\n",
    "    error_fa2 = output - labels\n",
    "    error_fa1 = np.matmul(error_fa2,fw2.T)\n",
    "    error_fa1 = np.multiply(error_fa1,fa1)\n",
    "    error_fa1 = np.multiply(error_fa1,(1-fa1))\n",
    "    error_temp = np.matmul(error_fa1,fw1.T)\n",
    "    error_temp = np.multiply(error_temp,temp_act)\n",
    "    error_temp = np.multiply(error_temp,(1-temp_act))\n",
    "    \n",
    "    m = dataset.shape[0]\n",
    "    \n",
    "    #cal errors conv layers\n",
    "    error_ca2 = np.array(error_temp).reshape(-1,image_size//4-1,image_size//4-1,depth*4)\n",
    "    error_ca1 = np.zeros(ca1.shape)\n",
    "    ## Image by Image error\n",
    "    for i in range(m):\n",
    "        error = conv_delta(error_ca2[i],cw2)\n",
    "        error = np.multiply(error,ca1[i])\n",
    "        error = np.multiply(error,(1-ca1[i]))\n",
    "        error_ca1 += error\n",
    "    \n",
    "    \n",
    "    #calculate partial derivatives\n",
    "    #fully connected layers\n",
    "    fd2 = (np.matmul(fa1.T,error_fa2) + beta*fw2)/m\n",
    "    fd1 = (np.matmul(temp_act.T,error_fa1) + beta*fw1)/m\n",
    "    \n",
    "    #conv layers\n",
    "    cd2 = np.zeros(cw2.shape)\n",
    "    cd1 = np.zeros(cw1.shape)\n",
    "    \n",
    "    ##Image by Image derivatives\n",
    "    for i in range(m):\n",
    "        cd2 = cd2 + conv_derivatives(error_ca2[i],ca1[i])\n",
    "        cd1 = cd1 + conv_derivatives(error_ca1[i],dataset[i])\n",
    "    cd2 = (cd2 + beta*cw2)/m\n",
    "    cd1 = (cd1 + beta*cw1)/m\n",
    "    \n",
    "    \n",
    "    #store the derivatives in dict\n",
    "    derivatives = dict()\n",
    "    \n",
    "    derivatives['cd1'] = cd1\n",
    "    derivatives['cd2'] = cd2\n",
    "    derivatives['fd1'] = fd1\n",
    "    derivatives['fd2'] = fd2\n",
    "    \n",
    "    return derivatives"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Update Parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def update_parameters(derivatives,parameters):\n",
    "    #get parameters\n",
    "    cw1 = parameters['cw1']\n",
    "    cw2 = parameters['cw2']\n",
    "    fw1 = parameters['fw1']\n",
    "    fw2 = parameters['fw2']\n",
    "    \n",
    "    #get derivatives\n",
    "    cd1 = derivatives['cd1']\n",
    "    cd2 = derivatives['cd2']\n",
    "    fd1 = derivatives['fd1']\n",
    "    fd2 = derivatives['fd2']\n",
    "    \n",
    "    #update\n",
    "    cw1 = cw1 - learning_rate*cd1\n",
    "    cw2 = cw2 - learning_rate*cd2\n",
    "    fw1 = fw1 - learning_rate*fd1\n",
    "    fw2 = fw2 - learning_rate*fd2\n",
    "    \n",
    "    #update the dict\n",
    "    parameters['cw1'] = cw1\n",
    "    parameters['cw2'] = cw2\n",
    "    parameters['fw1'] = fw1\n",
    "    parameters['fw2'] = fw2\n",
    "    \n",
    "    return parameters"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Loss and Accuracy\n",
    "* Loss = -sum((Y x log(predictions) + (1-Y) x (log(1-predictions))))/m + (beta x (sum(cw1^2) + sum(cw2^2) + sum(fw1^2) + sum(fw2^2)))\n",
    "* Accuray = sum(Y==Predictions)/m <br>\n",
    "\n",
    "m = number of examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "def cal_loss_accuracy(true_labels,predictions,parameters):\n",
    "    #get parameters\n",
    "    cw1 = parameters['cw1']\n",
    "    cw2 = parameters['cw2']\n",
    "    fw1 = parameters['fw1']\n",
    "    fw2 = parameters['fw2']\n",
    "    \n",
    "    m = len(true_labels)\n",
    "    \n",
    "    #cal loss\n",
    "    loss = -1*(np.sum(np.multiply(np.log(predictions),true_labels),1) + np.sum(np.multiply(np.log(1-predictions),1-true_labels),1))\n",
    "    loss = np.sum(loss)\n",
    "    loss = loss + beta*(np.sum(cw1**2) + np.sum(cw2**2) + np.sum(fw1**2) + np.sum(fw2**2))\n",
    "    loss = loss/m\n",
    "    \n",
    "    #cal accuracy\n",
    "    accuracy = np.sum(np.argmax(true_labels,1)==np.argmax(predictions,1))/m\n",
    "    \n",
    "    return loss,accuracy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train Function\n",
    "\n",
    "1. Initialize the parameters\n",
    "2. Forward Propagation\n",
    "3. Calculate Loss and Accuracy\n",
    "4. Backward Propagation\n",
    "5. Update the Parameters \n",
    "\n",
    "Batch Size = 20 <br>\n",
    "Repeat the steps 2-5 for every Batch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#train function\n",
    "def train(train_dataset,train_labels,batch_size=20,iters=101,stride=2):\n",
    "    \n",
    "    #initialize the parameters\n",
    "    parameters = initialize_parameters()\n",
    "    \n",
    "    cw1 = parameters['cw1']\n",
    "    cb1 = parameters['cb1']\n",
    "    cw2 = parameters['cw2']\n",
    "    cb2 = parameters['cb2']\n",
    "    \n",
    "    fw1 = parameters['fw1']\n",
    "    fb1 = parameters['fb1']\n",
    "    fw2 = parameters['fw2']\n",
    "    fb2 = parameters['fb2']\n",
    "    \n",
    "    J = []  #store the loss o every batch\n",
    "    A = []  #store the accuracy of every batch\n",
    "    \n",
    "    \n",
    "    #training process.\n",
    "    for step in range(iters):\n",
    "        #get the batch data.\n",
    "        start = (step*batch_size)%(train_dataset.shape[0])\n",
    "        end = start + batch_size\n",
    "        \n",
    "        batch_dataset = train_dataset[start:end,:,:,:]\n",
    "        batch_labels = train_labels[start:end,:]\n",
    "        \n",
    "        #forward propagation\n",
    "        cache,output = forward_propagation(batch_dataset,parameters)\n",
    "        \n",
    "        #cal_loss and accuracy\n",
    "        loss,accuracy = cal_loss_accuracy(batch_labels,output,parameters)\n",
    "        \n",
    "        #calculate the derivatives\n",
    "        derivatives = backward_propagation(batch_dataset,batch_labels,cache,parameters)\n",
    "        \n",
    "        #update the parameters\n",
    "        parameters = update_parameters(derivatives,parameters)\n",
    "        \n",
    "        #append the loss and accuracy of every batch\n",
    "        J.append(loss)\n",
    "        A.append(accuracy)\n",
    "        \n",
    "        #print loss and accuracy of the batch dataset.\n",
    "        if(step%100==0):\n",
    "            print('Step : %d'%step)\n",
    "            print('Loss : %f'%loss)\n",
    "            print('Accuracy : %f%%'%(round(accuracy*100,2)))\n",
    "            \n",
    "    return J,A,parameters"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###  Let's Train\n",
    "\n",
    "* 1501 Iterations is equivalent to ony one pass of entire training set (20*1500==30000).i.e an image is seen only once.\n",
    "* Takes 15-20 mins for 1501 iterations on CPU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Step : 0\n",
      "Loss : 29.588228\n",
      "Accuracy : 5.000000%\n",
      "Step : 100\n",
      "Loss : 3.916968\n",
      "Accuracy : 20.000000%\n",
      "Step : 200\n",
      "Loss : 1.254349\n",
      "Accuracy : 70.000000%\n",
      "Step : 300\n",
      "Loss : 1.120634\n",
      "Accuracy : 80.000000%\n",
      "Step : 400\n",
      "Loss : 0.926653\n",
      "Accuracy : 85.000000%\n",
      "Step : 500\n",
      "Loss : 1.102221\n",
      "Accuracy : 75.000000%\n",
      "Step : 600\n",
      "Loss : 1.095985\n",
      "Accuracy : 85.000000%\n",
      "Step : 700\n",
      "Loss : 0.311776\n",
      "Accuracy : 100.000000%\n",
      "Step : 800\n",
      "Loss : 0.874659\n",
      "Accuracy : 85.000000%\n",
      "Step : 900\n",
      "Loss : 0.298078\n",
      "Accuracy : 95.000000%\n",
      "Step : 1000\n",
      "Loss : 0.442242\n",
      "Accuracy : 95.000000%\n",
      "Step : 1100\n",
      "Loss : 0.881340\n",
      "Accuracy : 80.000000%\n",
      "Step : 1200\n",
      "Loss : 0.437191\n",
      "Accuracy : 90.000000%\n",
      "Step : 1300\n",
      "Loss : 0.764755\n",
      "Accuracy : 85.000000%\n",
      "Step : 1400\n",
      "Loss : 0.920038\n",
      "Accuracy : 75.000000%\n",
      "Step : 1500\n",
      "Loss : 0.678187\n",
      "Accuracy : 80.000000%\n"
     ]
    }
   ],
   "source": [
    "#TRAINING\n",
    "J,A,parameters = train(train_dataset,neural_train_labels,iters=1501)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "** The above accuracy and loss are of random batch data of size = 20, and not of entire training set **"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Let's Calculate the accuracy of entire Training Set, Valid Set and Test Set\n",
    "* Will take around 25-30 mins on CPU!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "#for training set\n",
    "_,train_pred = forward_propagation(train_dataset,parameters)\n",
    "_,train_accuracy = cal_loss_accuracy(neural_train_labels,train_pred,parameters)\n",
    "\n",
    "#for valid set\n",
    "_,valid_pred = forward_propagation(valid_dataset,parameters)\n",
    "_,valid_accuracy = cal_loss_accuracy(neural_valid_labels,valid_pred,parameters)\n",
    "\n",
    "#for test set\n",
    "_,test_pred = forward_propagation(test_dataset,parameters)\n",
    "_,test_accuracy = cal_loss_accuracy(neural_test_labels,test_pred,parameters)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Train Set = 87.94\n",
      "Accuracy of Valid Set = 87.77\n",
      "Accuracy of Test  Set = 87.88\n"
     ]
    }
   ],
   "source": [
    "#print\n",
    "print('Accuracy of Train Set = {}'.format(round(train_accuracy*100,2)))\n",
    "print('Accuracy of Valid Set = {}'.format(round(valid_accuracy*100,2)))\n",
    "print('Accuracy of Test  Set = {}'.format(round(test_accuracy*100,2)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "** So we have achieved an accuracy of 87% for all three datasets. Pretty Good!**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Let's Plot some graphs\n",
    "\n",
    "*  Plotted average loss of 30 batches and average accuracy of 30 batches."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAEWCAYAAABliCz2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nFxFJEgp8EZEkocAXEUkS/w9CARvVgRsEBgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x2078229edd8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEWCAYAAACJ0YulAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nHfM4/K5OKQUAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x2078056db00>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "avg_loss = list()\n",
    "avg_acc = list()\n",
    "i = 0\n",
    "while(i<len(J)):\n",
    "    avg_loss.append(np.mean(J[i:i+30]))\n",
    "    avg_acc.append(np.mean(A[i:i+30]))\n",
    "    i += 30\n",
    "\n",
    "plt.plot(list(range(len(avg_loss))),avg_loss)\n",
    "plt.xlabel(\"x\")\n",
    "plt.ylabel(\"Loss (Avg of 30 batches)\")\n",
    "plt.title(\"Loss Graph\")\n",
    "plt.show()\n",
    "\n",
    "plt.plot(list(range(len(avg_acc))),avg_acc)\n",
    "plt.xlabel(\"x\")\n",
    "plt.ylabel(\"Accuracy (Avg of 30 batches)\")\n",
    "plt.title(\"Accuracy Graph\")\n",
    "plt.show()    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Let's  visualize what the intermediate conv layers are learning!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Image\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAADOxJREFUeJzt3V+oXfWZxvHnmUyLmkbQNhNCGrWKDJRzkcohDIyUlsbg\nxMbYC7VBQ4SQFKmlhV74Zy7MlQRpG3IVOCWhcYi2kbYYtDgmYcAODsFjOOP/JrEkNOGYk2o1iQRa\nzduLsyKnevba273W2msn7/cDh7P3etef141Pfmvvtc7+OSIEIJ9/arsBAO0g/EBShB9IivADSRF+\nICnCDyRF+IGkCD+QFOEHkvrnQR7MNrcTAg2LCPeyXqWR3/bNtv9g+7DtB6rsC8Bgud97+23PkXRQ\n0k2Sjkl6UdLqiHi9ZBtGfqBhgxj5l0o6HBF/jIi/SvqlpFUV9gdggKqEf5GkP814fqxY9g9sb7A9\nbnu8wrEA1KzxD/wiYkzSmMRpPzBMqoz8xyUtnvH8y8UyABeAKuF/UdL1tr9i+/OSvitpdz1tAWha\n36f9EfGh7fsk/bekOZK2R8RrtXUGoFF9X+rr62C85wcaN5CbfABcuAg/kBThB5Ii/EBShB9IivAD\nSRF+ICnCDyRF+IGkCD+QFOEHkiL8QFKEH0iK8ANJEX4gKcIPJEX4gaQIP5AU4QeSIvxAUoQfSIrw\nA0kRfiApwg8kRfiBpAg/kBThB5Ii/EBShB9Iqu8puiXJ9hFJpyV9JOnDiBitoykAzasU/sI3I+LP\nNewHwABx2g8kVTX8IWmv7Zdsb6ijIQCDUfW0/8aIOG77XyTtsf1mRDw/c4XiHwX+YQCGjCOinh3Z\nGyWdiYiflKxTz8EAdBQR7mW9vk/7bc+1Pe/8Y0nLJb3a7/4ADFaV0/4Fkn5r+/x+Ho+IZ2vpCkDj\najvt7+lgSU/7H3zwwdL6I488Ulp//PHHS+t33XXXZ+5pGCxfvry0/uyz5WPJM888U1pfuXLlZ+7p\nYtD4aT+ACxvhB5Ii/EBShB9IivADSRF+IKk6/qoPXVx22WWl9W6XW8+cOVNnO0Pjuuuuq7R9t0uF\nN9xwQ8fagQMHKh37YsDIDyRF+IGkCD+QFOEHkiL8QFKEH0iK8ANJcZ1/AG6//fZK209MTNTUyXCp\nep3/7NmzpfVTp05V2v/FjpEfSIrwA0kRfiApwg8kRfiBpAg/kBThB5LiOn8NLr/88tL6pZdeWmn/\nJ0+erLR9m8rucVizZk2lfU9OTpbWDx8+XGn/FztGfiApwg8kRfiBpAg/kBThB5Ii/EBShB9Iqut1\nftvbJX1b0lREjBTLrpT0K0nXSDoi6Y6I+EtzbQ63kZGR0vrixYsr7f/gwYOVtm/SJZdcUlpfv359\nx9r8+fMrHbvb3/OjXC8j/y8k3fyJZQ9I2hcR10vaVzwHcAHpGv6IeF7Su59YvErSjuLxDkm31dwX\ngIb1+55/QUScv7fybUkLauoHwIBUvrc/IsJ2x8nmbG+QtKHqcQDUq9+R/4TthZJU/J7qtGJEjEXE\naESM9nksAA3oN/y7Ja0tHq+V9FQ97QAYlK7ht/2EpP+T9K+2j9leJ2mTpJtsH5K0rHgO4ALS9T1/\nRKzuUPpWzb2gg0OHDrXdQkePPvpoaX3ZsmWNHXvXrl2N7TsD7vADkiL8QFKEH0iK8ANJEX4gKcIP\nJMVXd9fg7rvvbruFxjz88MOl9XvvvbexY7///vul9e3btzd27AwY+YGkCD+QFOEHkiL8QFKEH0iK\n8ANJEX4gKa7z12DOnDltt9C3bvco3H///aX1Jv/bX3jhhdL61FTHL5BCDxj5gaQIP5AU4QeSIvxA\nUoQfSIrwA0kRfiAprvPXYGJiorR++vTp0vq8efNK61dffXVp/c033+xYW7RoUem2W7duLa13m4K7\nSUePHm3t2Bkw8gNJEX4gKcIPJEX4gaQIP5AU4QeSIvxAUo6I8hXs7ZK+LWkqIkaKZRslrZd0sljt\noYj4XdeD2eUHu0jt2LGjtL5mzZrS+pNPPlla37NnT8fa5s2bS7edO3duab1J586dK62vWLGitP7c\nc8/V2c5FIyLcy3q9jPy/kHTzLMs3R8SS4qdr8AEMl67hj4jnJb07gF4ADFCV9/w/sP2y7e22r6it\nIwAD0W/4t0q6VtISSZOSftppRdsbbI/bHu/zWAAa0Ff4I+JERHwUEeck/VzS0pJ1xyJiNCJG+20S\nQP36Cr/thTOefkfSq/W0A2BQuv5Jr+0nJH1D0pdsH5P0sKRv2F4iKSQdkfS9BnsE0ICu1/lrPVjS\n6/zLli0rrd93332l9ZUrV5bW7Z4u687q7NmzpfXdu3eX1u+8886+jz0+Xv4x0NKlHd9NokSd1/kB\nXIQIP5AU4QeSIvxAUoQfSIrwA0nx1d0DsHfv3kr1devWldZvvfXWjrVuX3+9ZcuW0vott9xSWq9y\nqW///v19b4vqGPmBpAg/kBThB5Ii/EBShB9IivADSRF+ICmu818Atm3bVqlexT333NPYvt97773G\n9o3uGPmBpAg/kBThB5Ii/EBShB9IivADSRF+ICmu86PU008/XVpfsmRJaf2tt97qWNu0aVNfPaEe\njPxAUoQfSIrwA0kRfiApwg8kRfiBpAg/kFTX6/y2F0t6TNICSSFpLCK22L5S0q8kXSPpiKQ7IuIv\nzbWKNoyMjFTavmwK8A8++KDSvlFNLyP/h5J+HBFflfRvkr5v+6uSHpC0LyKul7SveA7gAtE1/BEx\nGREHisenJb0haZGkVZJ2FKvtkHRbU00CqN9nes9v+xpJX5O0X9KCiJgsSm9r+m0BgAtEz/f22/6C\npF9L+lFEnLL9cS0iwnZ02G6DpA1VGwVQr55Gftuf03Twd0bEb4rFJ2wvLOoLJU3Ntm1EjEXEaESM\n1tEwgHp0Db+nh/htkt6IiJ/NKO2WtLZ4vFbSU/W3B6ApvZz2/7ukNZJesT1RLHtI0iZJu2yvk3RU\n0h3NtIg2vfPOO5W237VrV02doG5dwx8R/yvJHcrfqrcdAIPCHX5AUoQfSIrwA0kRfiApwg8kRfiB\npPjqbpS66qqrKm1f9ie9aBcjP5AU4QeSIvxAUoQfSIrwA0kRfiApwg8kxXV+lJo/f37bLaAhjPxA\nUoQfSIrwA0kRfiApwg8kRfiBpAg/kBTX+VHqzJkzbbeAhjDyA0kRfiApwg8kRfiBpAg/kBThB5Ii\n/EBSXa/z214s6TFJCySFpLGI2GJ7o6T1kk4Wqz4UEb9rqlG0Y/Xq1aX1nTt3DqgT1K2Xm3w+lPTj\niDhge56kl2zvKWqbI+InzbUHoCldwx8Rk5Imi8enbb8haVHTjQFo1md6z2/7Gklfk7S/WPQD2y/b\n3m77ig7bbLA9bnu8UqcAatVz+G1/QdKvJf0oIk5J2irpWklLNH1m8NPZtouIsYgYjYjRGvoFUJOe\nwm/7c5oO/s6I+I0kRcSJiPgoIs5J+rmkpc21CaBuXcNv25K2SXojIn42Y/nCGat9R9Kr9bcHoCmO\niPIV7Bsl/V7SK5LOFYsfkrRa06f8IemIpO8VHw6W7av8YAAqiwj3sl7X8NeJ8APN6zX83OEHJEX4\ngaQIP5AU4QeSIvxAUoQfSIrwA0kRfiApwg8kRfiBpAg/kBThB5Ii/EBShB9IatBTdP9Z0tEZz79U\nLBtGw9rbsPYl0Vu/6uzt6l5XHOjf83/q4Pb4sH6337D2Nqx9SfTWr7Z647QfSIrwA0m1Hf6xlo9f\nZlh7G9a+JHrrVyu9tfqeH0B72h75AbSklfDbvtn2H2wftv1AGz10YvuI7VdsT7Q9xVgxDdqU7Vdn\nLLvS9h7bh4rfs06T1lJvG20fL167CdsrWuptse3/sf267dds/7BY3uprV9JXK6/bwE/7bc+RdFDS\nTZKOSXpR0uqIeH2gjXRg+4ik0Yho/Zqw7a9LOiPpsYgYKZY9KundiNhU/MN5RUTcPyS9bZR0pu2Z\nm4sJZRbOnFla0m2S7lGLr11JX3eohdetjZF/qaTDEfHHiPirpF9KWtVCH0MvIp6X9O4nFq+StKN4\nvEPT//MMXIfehkJETEbEgeLxaUnnZ5Zu9bUr6asVbYR/kaQ/zXh+TMM15XdI2mv7Jdsb2m5mFgtm\nzIz0tqQFbTYzi64zNw/SJ2aWHprXrp8Zr+vGB36fdmNELJH0H5K+X5zeDqWYfs82TJdrepq5eVBm\nmVn6Y22+dv3OeF23NsJ/XNLiGc+/XCwbChFxvPg9Jem3Gr7Zh0+cnyS1+D3Vcj8fG6aZm2ebWVpD\n8NoN04zXbYT/RUnX2/6K7c9L+q6k3S308Sm25xYfxMj2XEnLNXyzD++WtLZ4vFbSUy328g+GZebm\nTjNLq+XXbuhmvI6Igf9IWqHpT/zfkvSfbfTQoa9rJf1/8fNa271JekLTp4F/0/RnI+skfVHSPkmH\nJO2VdOUQ9fZfmp7N+WVNB21hS73dqOlT+pclTRQ/K9p+7Ur6auV14w4/ICk+8AOSIvxAUoQfSIrw\nA0kRfiApwg8kRfiBpAg/kNTfAXp7DvVSJ/viAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x207822a2dd8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "True_Label = 4\n",
      "Pred_Label = 4\n"
     ]
    }
   ],
   "source": [
    "index = 10\n",
    "test_image = test_dataset[index].reshape(1,28,28,1)\n",
    "plt.imshow(test_image[0,:,:,0],cmap='gray')\n",
    "print(\"Image\")\n",
    "plt.show()\n",
    "print(\"True_Label = {}\".format(np.argmax(neural_test_labels[index])))\n",
    "print(\"Pred_Label = {}\".format(np.argmax(test_pred[index])))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cache,pred = forward_propagation(test_image,parameters)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "conv_layer1 = cache['ca1']\n",
    "conv_layer2 = cache['ca2']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Conv layer1 Activations\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x20790817f60>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nB5Ki/EBSlB9IivIDSVF+ICnKDyT1X+iHnjrSGM5mAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x207905cad68>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n8QklAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x207908b47f0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x20782358ba8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Conv layer2 Activations\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAW4AAABLCAYAAABQtG2+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x20790817e48>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(\"Conv layer1 Activations\")\n",
    "for i in range(conv_layer1.shape[3]):\n",
    "    plt.imshow(conv_layer1[0,:,:,i],cmap='gray')\n",
    "    plt.show()\n",
    "\n",
    "fig2 = plt.figure()\n",
    "print(\"Conv layer2 Activations\")\n",
    "for i in range(0,(conv_layer2.shape[3]-9)):\n",
    "    fig2.add_subplot(1,7,i+1)\n",
    "    plt.imshow(conv_layer2[0,:,:,i],cmap='gray')\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Some Predictions made by the MODEL  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n/FJThl9qyvBLTf0/eTLL1Gx+KuEAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x207823a9940>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted Label = 1\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x207906f3f60>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted Label = 5\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nFWX4paIMv1SU4ZeKMvxSUf8HFW1EjcsXbR8AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x207906ca898>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted Label = 9\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nQFCEHwiK8ANBEX4gKMIPBPV/NoWUAuJ9RNMAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x2079084aba8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted Label = 0\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x2078465d198>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicted Label = 2\n"
     ]
    }
   ],
   "source": [
    "images = test_dataset[0:5]\n",
    "image_labels = neural_test_labels[0:5]\n",
    "\n",
    "for i in range(len(images)):\n",
    "    plt.imshow(images[i,:,:,0],cmap='gray')\n",
    "    plt.show()\n",
    "    print(\"Predicted Label = {}\".format(np.argmax(image_labels[i])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# THE END"
   ]
  }
 ],
 "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": "3.9.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}