04. Neural Networks/NeuralNetworks.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "NeuralNetworks.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "source": [
        "**INITIALIZATION:**"
      ],
      "metadata": {
        "id": "4kMIRQvmkig0"
      }
    },
    {
      "cell_type": "code",
      "execution_count": 16,
      "metadata": {
        "id": "AlCfmSxgNqhr"
      },
      "outputs": [],
      "source": [
        "#@ INITIALIZING NECESSARY PACKAGES AND DEPENDENCIES: \n",
        "from sklearn.preprocessing import LabelBinarizer\n",
        "from sklearn.model_selection import train_test_split\n",
        "from sklearn.metrics import classification_report\n",
        "from sklearn import datasets\n",
        "\n",
        "from keras.models import Sequential \n",
        "from keras.layers.core import Dense\n",
        "from keras.datasets import cifar10\n",
        "from tensorflow.keras.optimizers import SGD\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**PERCEPTRON:**"
      ],
      "metadata": {
        "id": "2hpm7_bOkc1o"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@ INITIALIZING PERCEPTRON MODEL: \n",
        "class Perceptron:                                       # Defining Perceptron. \n",
        "    def __init__(self, N, alpha=0.1):                   # Initializing Constructor Function. \n",
        "        self.W = np.random.randn(N + 1) / np.sqrt(N)    # Initializing Scaled Weight Matrix. \n",
        "        self.alpha = alpha                              # Initializing LR.\n",
        "    \n",
        "    def step(self, x):                                  # Defining Step Function. \n",
        "        return 1 if x > 0 else 0                        # Getting 1 if Positive else Negative. \n",
        "    \n",
        "    def fit(self, X, y, epochs=10):                     # Defining Fit Function. \n",
        "        X = np.c_[X, np.ones((X.shape[0]))]             # Adding Column of Ones. \n",
        "        for epoch in np.arange(0, epochs):\n",
        "            for (x, target) in zip(X, y):\n",
        "                p = self.step(np.dot(x, self.W))        # Initializing Dot Product. \n",
        "                if p != target:\n",
        "                    error = p - target                  # Computing Error. \n",
        "                    self.W += -self.alpha*error*x       # Updating Weight Matrix. \n",
        "    \n",
        "    def predict(self, X, addBias=True):                 # Defining Predict Function. \n",
        "        X = np.atleast_2d(X)                            # Inspecting 2D Matrix. \n",
        "        if addBias:\n",
        "            X = np.c_[X, np.ones((X.shape[0]))]         # Adding Bias Vector. \n",
        "        return self.step(np.dot(X, self.W))             # Initializing Dot Product. "
      ],
      "metadata": {
        "id": "14phxR8kRMtB"
      },
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#@ EVALUATING PERCEPTRON BITWISE DATASETS: OR:\n",
        "X = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])          # Initializing Array Example. \n",
        "y = np.array([[0], [1], [1], [1]])                      # Initializing Array Example. \n",
        "\n",
        "#@ TRAINING PERCEPTRON MODEL: \n",
        "p = Perceptron(X.shape[1], alpha=0.1)                   # Initializing Perceptron Model. \n",
        "p.fit(X, y, epochs=20)                                  # Training Model. \n",
        "\n",
        "#@ MODEL EVALUATION: \n",
        "for (x, target) in zip(X, y):\n",
        "    pred = p.predict(x)\n",
        "    print(f\"data={x}, ground-truth={target}, pred={pred}\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "YqHQfynQb_G_",
        "outputId": "e6561325-c603-4ba5-ec9e-c6e9f929d4e3"
      },
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "data=[0 0], ground-truth=[0], pred=0\n",
            "data=[0 1], ground-truth=[1], pred=1\n",
            "data=[1 0], ground-truth=[1], pred=1\n",
            "data=[1 1], ground-truth=[1], pred=1\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**NEURAL NETWORKS: BACKPROPAGATION**"
      ],
      "metadata": {
        "id": "vCm97EO6rjta"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@ INITIALIZING BACKPROPAGATION ALGORITHM:\n",
        "class NeuralNetwork:                                            # Defining Neural Network. \n",
        "    def __init__(self, layers, alpha=0.1):                      # Initializing Constructor Function. \n",
        "        self.W = []                                             # Initialization. \n",
        "        self.layers = layers                                    # Initialization. \n",
        "        self.alpha = alpha                                      # Initialization. \n",
        "        for i in np.arange(0, len(layers) - 2):\n",
        "            w = np.random.randn(layers[i]+1, layers[i+1]+1)     # Initializing Weight Matrix. \n",
        "            self.W.append(w / np.sqrt(layers[i]))               # Normalizing Variance. \n",
        "        w = np.random.randn(layers[-2]+1, layers[-1])           # Initializing Weight Matrix. \n",
        "        self.W.append(w / np.sqrt(layers[-2]))                  # Normalizing Variance. \n",
        "    \n",
        "    def __repr__(self):                                         # Function for Debugging. \n",
        "        return \"NeuralNetwork: {}\".format(\n",
        "            \"-\".join(str(l) for l in self.layers))              # Inspecting. \n",
        "    \n",
        "    def sigmoid(self, x):                                       # Defining Sigmoid Function. \n",
        "        return 1.0 / (1 + np.exp(-x))                           # Implementation of Sigmoid Function. \n",
        "    \n",
        "    def sigmoid_deriv(self, x):                                 # Defining Function. \n",
        "        return x * (1 - x)                                      # Getting Derivative of Sigmoid. \n",
        "    \n",
        "    def fit(self, X, y, epochs=1000, displayUpdate=100):        # Defining Fit Function. \n",
        "        X = np.c_[X, np.ones((X.shape[0]))]                     # Adding Column of Ones. \n",
        "        for epoch in np.arange(0, epochs):\n",
        "            for (x, target) in zip(X, y):\n",
        "                self.fit_partial(x, target)                     # Backpropagation and Updating Matrix. \n",
        "            if epoch==0 or (epoch+1)%displayUpdate==0:\n",
        "                loss = self.calculate_loss(X, y)                # Initializing Loss Calculation. \n",
        "                print(\"epoch={}, loss={:.7f}\".format(\n",
        "                    epoch+1, loss))                             # Inspecting Loss. \n",
        "    \n",
        "    def fit_partial(self, x, y):                                # Defining Fit Partial Method. \n",
        "        A = [np.atleast_2d(x)]                                  # Initializing List. \n",
        "\n",
        "        #@ FEEDFORWARD:\n",
        "        for layer in np.arange(0, len(self.W)):\n",
        "            net = A[layer].dot(self.W[layer])                   # Dot Product of Activation and Weight Matrix. \n",
        "            out = self.sigmoid(net)                             # Implementing Nonlinear Activation Function. \n",
        "            A.append(out)                                       # Adding to Activations. \n",
        "\n",
        "        #@ BACKPROPAGATION:\n",
        "        error = A[-1] - y                                       # Computing Error. \n",
        "        D = [error * self.sigmoid_deriv(A[-1])]                 # Initializing List of Deltas. \n",
        "        for layer in np.arange(len(A) - 2, 0, -1):\n",
        "            delta = D[-1].dot(self.W[layer].T)                  # Initializing Dot Product. \n",
        "            delta = delta * self.sigmoid_deriv(A[layer])        # Implementating Derivative Function. \n",
        "            D.append(delta)                                     # Adding Activations. \n",
        "        D = D[::-1]                                             # Reversing Activations. \n",
        "\n",
        "        #@ WEIGHT UPDATE PHASE: \n",
        "        for layer in np.arange(0, len(self.W)):\n",
        "            self.W[layer] += -self.alpha * A[layer].T.dot(D[layer])\n",
        "        \n",
        "    def predict(self, X, addBias=True):                         # Defining Predict Function. \n",
        "        p = np.atleast_2d(X)                                    # Initialization. \n",
        "        if addBias:\n",
        "            p = np.c_[p, np.ones((p.shape[0]))]                 # Adding Column of Ones. \n",
        "        for layer in np.arange(0, len(self.W)):\n",
        "            p = self.sigmoid(np.dot(p, self.W[layer]))          # Implementing Dot Product. \n",
        "        return p\n",
        "    \n",
        "    def calculate_loss(self, X, targets):                       # Defining Function. \n",
        "        targets = np.atleast_2d(targets)                        # Initialization. \n",
        "        predictions = self.predict(X, addBias=False)            # Implementation of Predict Method. \n",
        "        loss = 0.5 * np.sum((predictions - targets)**2)         # Computing Loss. \n",
        "        return loss"
      ],
      "metadata": {
        "id": "wgKuObXhri8_"
      },
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#@ EVALUATING NEURAL NETWORKS ON BITWISE DATASETS: XOR:\n",
        "X = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])                  # Initializing Array Example. \n",
        "y = np.array([[0], [1], [1], [0]])                              # Initializing Array Example. \n",
        "\n",
        "#@ TRAINING NEURAL NETWORKS MODEL: \n",
        "nn = NeuralNetwork([2, 2, 1], alpha=0.5)                        # Initializing Neural Network.  \n",
        "nn.fit(X, y, epochs=200)                                        # Training Model. \n",
        "\n",
        "#@ MODEL EVALUATION: \n",
        "for (x, target) in zip(X, y):\n",
        "    pred = nn.predict(x)[0][0]\n",
        "    step = 1 if pred > 0.5 else 0\n",
        "    print(f\"data={x}, ground-truth={target[0]}, pred={pred},\"\n",
        "          f\"step={step}\")"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "elrPa9oaNVr7",
        "outputId": "e892ed17-b610-4bca-9c37-5485803e6b80"
      },
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch=1, loss=0.5317512\n",
            "epoch=100, loss=0.4957142\n",
            "epoch=200, loss=0.4913100\n",
            "data=[0 0], ground-truth=0, pred=0.4751030513080377,step=0\n",
            "data=[0 1], ground-truth=1, pred=0.47896024764804435,step=0\n",
            "data=[1 0], ground-truth=1, pred=0.5279040479362642,step=1\n",
            "data=[1 1], ground-truth=0, pred=0.5123867348849976,step=1\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**MNIST SAMPLE:**"
      ],
      "metadata": {
        "id": "oeV74YDzh1PC"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@ GETTING THE DATASET:\n",
        "digits = datasets.load_digits()                         # Initializing Dataset. \n",
        "data = digits.data.astype(\"float\")                      # Converting into Float. \n",
        "data = (data - data.min()) / (data.max() - data.min())  # Normalizing Dataset. \n",
        "print(\"samples: {}, dim: {}\".format(data.shape[0],\n",
        "                                  data.shape[1]))       # Inspecting Dataset. "
      ],
      "metadata": {
        "id": "fN8n83WbP1G_",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "c09a8119-0141-439e-f4fb-346ad06e1cb4"
      },
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "samples: 1797, dim: 64\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#@ PREPARING THE DATASET:\n",
        "(train_X, test_X, train_Y, test_Y) = train_test_split(\n",
        "    data, digits.target, test_size=0.25)                    # Splitting Dataset. \n",
        "train_Y = LabelBinarizer().fit_transform(train_Y)           # Converting Labels to Vectors. \n",
        "test_Y = LabelBinarizer().fit_transform(test_Y)             # Converting Labels to Vectors. \n",
        "\n",
        "#@ TRAINING THE NEURAL NETWORK:\n",
        "nn = NeuralNetwork([train_X.shape[1], 32, 16, 10])          # Initializing Neural Network. \n",
        "nn.fit(train_X, train_Y, epochs=1000)                       # Training Neural Network.\n",
        " \n",
        "#@ EVALUATING THE NEURAL NETWORK: \n",
        "predictions = nn.predict(test_X)                            # Getting Predictions. \n",
        "predictions = predictions.argmax(axis=1)                    # Getting Maximum Probability. \n",
        "print(classification_report(\n",
        "    test_Y.argmax(axis=1), predictions))                    # Inspecting Classification Report. "
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "dGX1TqUvj6vy",
        "outputId": "95213399-08ec-461a-e975-d33c64a81708"
      },
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch=1, loss=603.6436047\n",
            "epoch=100, loss=6.7991431\n",
            "epoch=200, loss=2.3439244\n",
            "epoch=300, loss=1.5595649\n",
            "epoch=400, loss=1.3600940\n",
            "epoch=500, loss=1.2639233\n",
            "epoch=600, loss=1.2070289\n",
            "epoch=700, loss=1.1679645\n",
            "epoch=800, loss=0.6600999\n",
            "epoch=900, loss=0.6320372\n",
            "epoch=1000, loss=0.6139866\n",
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.96      1.00      0.98        44\n",
            "           1       0.95      0.91      0.93        57\n",
            "           2       0.96      0.98      0.97        48\n",
            "           3       1.00      0.95      0.98        42\n",
            "           4       1.00      1.00      1.00        41\n",
            "           5       0.96      0.96      0.96        53\n",
            "           6       1.00      0.97      0.99        37\n",
            "           7       0.98      0.98      0.98        44\n",
            "           8       0.86      0.93      0.89        41\n",
            "           9       0.98      0.95      0.96        43\n",
            "\n",
            "    accuracy                           0.96       450\n",
            "   macro avg       0.96      0.96      0.96       450\n",
            "weighted avg       0.96      0.96      0.96       450\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**NEURAL NETWORKS:**"
      ],
      "metadata": {
        "id": "Z_QhpG32tp6g"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@ GETTING THE DATASET: \n",
        "dataset = datasets.fetch_openml(\"mnist_784\")                # Downloading Dataset. \n",
        "data = dataset.data.astype(\"float\") / 255.0                 # Initializing Data Normalization. \n",
        "\n",
        "#@ PREPARING THE DATASET:\n",
        "lb = LabelBinarizer()                                       # Initializing LabelBinarizer. \n",
        "(train_X, test_X, train_Y, test_Y) = train_test_split(\n",
        "    data, dataset.target, test_size=0.25)                   # Splitting Dataset. \n",
        "train_Y = LabelBinarizer().fit_transform(train_Y)           # Converting Labels to Vectors. \n",
        "test_Y = LabelBinarizer().fit_transform(test_Y)             # Converting Labels to Vectors. "
      ],
      "metadata": {
        "id": "YPSRJZIUs_nU"
      },
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#@ DEFINING NETWORK ARCHITECTURE: \n",
        "model = Sequential()                                            # Initializing Sequential Model. \n",
        "model.add(Dense(256, input_shape=(784,), activation=\"sigmoid\")) # Initializing Dense Layer. \n",
        "model.add(Dense(128, activation=\"sigmoid\"))                     # Adding Dense Layer. \n",
        "model.add(Dense(10, activation=\"softmax\"))                      # Adding Dense Output Layer. \n",
        "\n",
        "#@ TRAINING NEURAL NETWORK:\n",
        "sgd = SGD(0.01)                                                 # Initializing SGD Optimizer. \n",
        "model.compile(loss=\"categorical_crossentropy\", optimizer=sgd, \n",
        "              metrics=[\"accuracy\"])                             # Compiling Model. \n",
        "H = model.fit(train_X, train_Y,validation_data=(test_X,test_Y), \n",
        "              epochs=100, batch_size=128)                       # Training Model. "
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "FMQacNBtv6Pr",
        "outputId": "583e81d4-44a5-4904-9246-dc9ca9638f9b"
      },
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/100\n",
            "411/411 [==============================] - 6s 12ms/step - loss: 2.2737 - accuracy: 0.2045 - val_loss: 2.2423 - val_accuracy: 0.3751\n",
            "Epoch 2/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 2.2136 - accuracy: 0.3976 - val_loss: 2.1813 - val_accuracy: 0.4214\n",
            "Epoch 3/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 2.1438 - accuracy: 0.5190 - val_loss: 2.1013 - val_accuracy: 0.5834\n",
            "Epoch 4/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 2.0489 - accuracy: 0.5854 - val_loss: 1.9892 - val_accuracy: 0.6376\n",
            "Epoch 5/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 1.9176 - accuracy: 0.6274 - val_loss: 1.8358 - val_accuracy: 0.6569\n",
            "Epoch 6/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 1.7459 - accuracy: 0.6600 - val_loss: 1.6478 - val_accuracy: 0.6816\n",
            "Epoch 7/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 1.5493 - accuracy: 0.6950 - val_loss: 1.4484 - val_accuracy: 0.6896\n",
            "Epoch 8/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 1.3572 - accuracy: 0.7259 - val_loss: 1.2665 - val_accuracy: 0.7571\n",
            "Epoch 9/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 1.1914 - accuracy: 0.7540 - val_loss: 1.1168 - val_accuracy: 0.7749\n",
            "Epoch 10/100\n",
            "411/411 [==============================] - 5s 13ms/step - loss: 1.0578 - accuracy: 0.7755 - val_loss: 0.9982 - val_accuracy: 0.7809\n",
            "Epoch 11/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 0.9519 - accuracy: 0.7907 - val_loss: 0.9039 - val_accuracy: 0.7994\n",
            "Epoch 12/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 0.8674 - accuracy: 0.8034 - val_loss: 0.8283 - val_accuracy: 0.8127\n",
            "Epoch 13/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 0.7990 - accuracy: 0.8141 - val_loss: 0.7673 - val_accuracy: 0.8170\n",
            "Epoch 14/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 0.7430 - accuracy: 0.8226 - val_loss: 0.7158 - val_accuracy: 0.8286\n",
            "Epoch 15/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 0.6965 - accuracy: 0.8302 - val_loss: 0.6739 - val_accuracy: 0.8323\n",
            "Epoch 16/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 0.6576 - accuracy: 0.8367 - val_loss: 0.6381 - val_accuracy: 0.8402\n",
            "Epoch 17/100\n",
            "411/411 [==============================] - 6s 14ms/step - loss: 0.6247 - accuracy: 0.8425 - val_loss: 0.6075 - val_accuracy: 0.8456\n",
            "Epoch 18/100\n",
            "411/411 [==============================] - 5s 13ms/step - loss: 0.5964 - accuracy: 0.8482 - val_loss: 0.5813 - val_accuracy: 0.8509\n",
            "Epoch 19/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 0.5720 - accuracy: 0.8533 - val_loss: 0.5593 - val_accuracy: 0.8540\n",
            "Epoch 20/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.5506 - accuracy: 0.8568 - val_loss: 0.5391 - val_accuracy: 0.8598\n",
            "Epoch 21/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.5319 - accuracy: 0.8618 - val_loss: 0.5215 - val_accuracy: 0.8633\n",
            "Epoch 22/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 0.5152 - accuracy: 0.8644 - val_loss: 0.5061 - val_accuracy: 0.8669\n",
            "Epoch 23/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.5002 - accuracy: 0.8677 - val_loss: 0.4926 - val_accuracy: 0.8694\n",
            "Epoch 24/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.4868 - accuracy: 0.8703 - val_loss: 0.4798 - val_accuracy: 0.8722\n",
            "Epoch 25/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 0.4746 - accuracy: 0.8732 - val_loss: 0.4688 - val_accuracy: 0.8746\n",
            "Epoch 26/100\n",
            "411/411 [==============================] - 5s 13ms/step - loss: 0.4636 - accuracy: 0.8758 - val_loss: 0.4588 - val_accuracy: 0.8778\n",
            "Epoch 27/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.4536 - accuracy: 0.8783 - val_loss: 0.4493 - val_accuracy: 0.8791\n",
            "Epoch 28/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.4445 - accuracy: 0.8803 - val_loss: 0.4403 - val_accuracy: 0.8816\n",
            "Epoch 29/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.4360 - accuracy: 0.8816 - val_loss: 0.4326 - val_accuracy: 0.8834\n",
            "Epoch 30/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.4284 - accuracy: 0.8831 - val_loss: 0.4263 - val_accuracy: 0.8846\n",
            "Epoch 31/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.4213 - accuracy: 0.8848 - val_loss: 0.4192 - val_accuracy: 0.8864\n",
            "Epoch 32/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.4147 - accuracy: 0.8867 - val_loss: 0.4130 - val_accuracy: 0.8875\n",
            "Epoch 33/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.4086 - accuracy: 0.8878 - val_loss: 0.4080 - val_accuracy: 0.8893\n",
            "Epoch 34/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 0.4030 - accuracy: 0.8891 - val_loss: 0.4024 - val_accuracy: 0.8890\n",
            "Epoch 35/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3977 - accuracy: 0.8903 - val_loss: 0.3974 - val_accuracy: 0.8898\n",
            "Epoch 36/100\n",
            "411/411 [==============================] - 4s 11ms/step - loss: 0.3928 - accuracy: 0.8916 - val_loss: 0.3932 - val_accuracy: 0.8913\n",
            "Epoch 37/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3882 - accuracy: 0.8926 - val_loss: 0.3890 - val_accuracy: 0.8914\n",
            "Epoch 38/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3839 - accuracy: 0.8932 - val_loss: 0.3848 - val_accuracy: 0.8924\n",
            "Epoch 39/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3798 - accuracy: 0.8943 - val_loss: 0.3815 - val_accuracy: 0.8931\n",
            "Epoch 40/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3759 - accuracy: 0.8950 - val_loss: 0.3776 - val_accuracy: 0.8943\n",
            "Epoch 41/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3724 - accuracy: 0.8957 - val_loss: 0.3742 - val_accuracy: 0.8954\n",
            "Epoch 42/100\n",
            "411/411 [==============================] - 6s 14ms/step - loss: 0.3689 - accuracy: 0.8968 - val_loss: 0.3712 - val_accuracy: 0.8965\n",
            "Epoch 43/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3656 - accuracy: 0.8974 - val_loss: 0.3680 - val_accuracy: 0.8959\n",
            "Epoch 44/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3625 - accuracy: 0.8985 - val_loss: 0.3650 - val_accuracy: 0.8972\n",
            "Epoch 45/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3596 - accuracy: 0.8990 - val_loss: 0.3628 - val_accuracy: 0.8986\n",
            "Epoch 46/100\n",
            "411/411 [==============================] - 5s 11ms/step - loss: 0.3567 - accuracy: 0.8999 - val_loss: 0.3599 - val_accuracy: 0.8981\n",
            "Epoch 47/100\n",
            "411/411 [==============================] - 4s 11ms/step - loss: 0.3540 - accuracy: 0.9006 - val_loss: 0.3571 - val_accuracy: 0.8990\n",
            "Epoch 48/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3513 - accuracy: 0.9010 - val_loss: 0.3548 - val_accuracy: 0.8993\n",
            "Epoch 49/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 0.3489 - accuracy: 0.9016 - val_loss: 0.3526 - val_accuracy: 0.8995\n",
            "Epoch 50/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 0.3464 - accuracy: 0.9020 - val_loss: 0.3506 - val_accuracy: 0.9003\n",
            "Epoch 51/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3441 - accuracy: 0.9025 - val_loss: 0.3486 - val_accuracy: 0.9016\n",
            "Epoch 52/100\n",
            "411/411 [==============================] - 4s 11ms/step - loss: 0.3419 - accuracy: 0.9035 - val_loss: 0.3463 - val_accuracy: 0.9014\n",
            "Epoch 53/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3397 - accuracy: 0.9039 - val_loss: 0.3442 - val_accuracy: 0.9026\n",
            "Epoch 54/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3375 - accuracy: 0.9042 - val_loss: 0.3424 - val_accuracy: 0.9020\n",
            "Epoch 55/100\n",
            "411/411 [==============================] - 5s 12ms/step - loss: 0.3355 - accuracy: 0.9044 - val_loss: 0.3407 - val_accuracy: 0.9034\n",
            "Epoch 56/100\n",
            "411/411 [==============================] - 6s 14ms/step - loss: 0.3335 - accuracy: 0.9055 - val_loss: 0.3385 - val_accuracy: 0.9042\n",
            "Epoch 57/100\n",
            "411/411 [==============================] - 4s 10ms/step - loss: 0.3315 - accuracy: 0.9060 - val_loss: 0.3371 - val_accuracy: 0.9043\n",
            "Epoch 58/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3296 - accuracy: 0.9068 - val_loss: 0.3357 - val_accuracy: 0.9044\n",
            "Epoch 59/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3278 - accuracy: 0.9072 - val_loss: 0.3335 - val_accuracy: 0.9049\n",
            "Epoch 60/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3261 - accuracy: 0.9077 - val_loss: 0.3317 - val_accuracy: 0.9055\n",
            "Epoch 61/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3243 - accuracy: 0.9078 - val_loss: 0.3303 - val_accuracy: 0.9055\n",
            "Epoch 62/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.3226 - accuracy: 0.9085 - val_loss: 0.3292 - val_accuracy: 0.9054\n",
            "Epoch 63/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3210 - accuracy: 0.9086 - val_loss: 0.3271 - val_accuracy: 0.9066\n",
            "Epoch 64/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.3193 - accuracy: 0.9092 - val_loss: 0.3256 - val_accuracy: 0.9066\n",
            "Epoch 65/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3178 - accuracy: 0.9094 - val_loss: 0.3241 - val_accuracy: 0.9066\n",
            "Epoch 66/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.3162 - accuracy: 0.9100 - val_loss: 0.3230 - val_accuracy: 0.9073\n",
            "Epoch 67/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3147 - accuracy: 0.9107 - val_loss: 0.3215 - val_accuracy: 0.9074\n",
            "Epoch 68/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3132 - accuracy: 0.9108 - val_loss: 0.3199 - val_accuracy: 0.9078\n",
            "Epoch 69/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3118 - accuracy: 0.9116 - val_loss: 0.3185 - val_accuracy: 0.9082\n",
            "Epoch 70/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.3104 - accuracy: 0.9118 - val_loss: 0.3176 - val_accuracy: 0.9078\n",
            "Epoch 71/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.3089 - accuracy: 0.9118 - val_loss: 0.3165 - val_accuracy: 0.9083\n",
            "Epoch 72/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3075 - accuracy: 0.9126 - val_loss: 0.3150 - val_accuracy: 0.9089\n",
            "Epoch 73/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3061 - accuracy: 0.9127 - val_loss: 0.3143 - val_accuracy: 0.9094\n",
            "Epoch 74/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.3049 - accuracy: 0.9133 - val_loss: 0.3128 - val_accuracy: 0.9093\n",
            "Epoch 75/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.3035 - accuracy: 0.9138 - val_loss: 0.3113 - val_accuracy: 0.9099\n",
            "Epoch 76/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3023 - accuracy: 0.9139 - val_loss: 0.3101 - val_accuracy: 0.9102\n",
            "Epoch 77/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.3010 - accuracy: 0.9142 - val_loss: 0.3092 - val_accuracy: 0.9101\n",
            "Epoch 78/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2998 - accuracy: 0.9146 - val_loss: 0.3079 - val_accuracy: 0.9098\n",
            "Epoch 79/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2986 - accuracy: 0.9150 - val_loss: 0.3066 - val_accuracy: 0.9112\n",
            "Epoch 80/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2973 - accuracy: 0.9154 - val_loss: 0.3056 - val_accuracy: 0.9108\n",
            "Epoch 81/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2962 - accuracy: 0.9157 - val_loss: 0.3048 - val_accuracy: 0.9105\n",
            "Epoch 82/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2951 - accuracy: 0.9158 - val_loss: 0.3034 - val_accuracy: 0.9122\n",
            "Epoch 83/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.2938 - accuracy: 0.9161 - val_loss: 0.3023 - val_accuracy: 0.9121\n",
            "Epoch 84/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2928 - accuracy: 0.9166 - val_loss: 0.3014 - val_accuracy: 0.9126\n",
            "Epoch 85/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.2917 - accuracy: 0.9171 - val_loss: 0.3004 - val_accuracy: 0.9127\n",
            "Epoch 86/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.2906 - accuracy: 0.9175 - val_loss: 0.2996 - val_accuracy: 0.9133\n",
            "Epoch 87/100\n",
            "411/411 [==============================] - 4s 9ms/step - loss: 0.2895 - accuracy: 0.9174 - val_loss: 0.2982 - val_accuracy: 0.9134\n",
            "Epoch 88/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2884 - accuracy: 0.9177 - val_loss: 0.2971 - val_accuracy: 0.9141\n",
            "Epoch 89/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.2874 - accuracy: 0.9180 - val_loss: 0.2961 - val_accuracy: 0.9145\n",
            "Epoch 90/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2863 - accuracy: 0.9185 - val_loss: 0.2954 - val_accuracy: 0.9143\n",
            "Epoch 91/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.2853 - accuracy: 0.9187 - val_loss: 0.2943 - val_accuracy: 0.9147\n",
            "Epoch 92/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2842 - accuracy: 0.9189 - val_loss: 0.2933 - val_accuracy: 0.9149\n",
            "Epoch 93/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2832 - accuracy: 0.9193 - val_loss: 0.2925 - val_accuracy: 0.9151\n",
            "Epoch 94/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2823 - accuracy: 0.9192 - val_loss: 0.2915 - val_accuracy: 0.9157\n",
            "Epoch 95/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.2813 - accuracy: 0.9198 - val_loss: 0.2907 - val_accuracy: 0.9158\n",
            "Epoch 96/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2803 - accuracy: 0.9199 - val_loss: 0.2902 - val_accuracy: 0.9161\n",
            "Epoch 97/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.2793 - accuracy: 0.9204 - val_loss: 0.2887 - val_accuracy: 0.9159\n",
            "Epoch 98/100\n",
            "411/411 [==============================] - 3s 7ms/step - loss: 0.2784 - accuracy: 0.9208 - val_loss: 0.2885 - val_accuracy: 0.9163\n",
            "Epoch 99/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2774 - accuracy: 0.9207 - val_loss: 0.2872 - val_accuracy: 0.9170\n",
            "Epoch 100/100\n",
            "411/411 [==============================] - 3s 8ms/step - loss: 0.2765 - accuracy: 0.9210 - val_loss: 0.2863 - val_accuracy: 0.9169\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#@ MODEL EVALUATION: \n",
        "predictions = model.predict(test_X, batch_size=128)             # Getting Predictions. \n",
        "print(classification_report(test_Y.argmax(axis=1), \n",
        "                            predictions.argmax(axis=1)))        # Inspecting Classification Report. "
      ],
      "metadata": {
        "id": "BjEIDydK5Vyx",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "f04fd919-eafb-4c2f-d5a2-01031934eafc"
      },
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "           0       0.95      0.96      0.96      1728\n",
            "           1       0.94      0.97      0.96      1938\n",
            "           2       0.91      0.90      0.91      1743\n",
            "           3       0.92      0.89      0.90      1798\n",
            "           4       0.91      0.93      0.92      1639\n",
            "           5       0.89      0.88      0.88      1608\n",
            "           6       0.94      0.96      0.95      1732\n",
            "           7       0.94      0.94      0.94      1849\n",
            "           8       0.90      0.87      0.89      1684\n",
            "           9       0.90      0.89      0.90      1781\n",
            "\n",
            "    accuracy                           0.92     17500\n",
            "   macro avg       0.92      0.92      0.92     17500\n",
            "weighted avg       0.92      0.92      0.92     17500\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#@ INSPECTING TRAINING LOSS AND ACCURACY:\n",
        "plt.style.use(\"ggplot\")\n",
        "plt.figure()\n",
        "plt.plot(np.arange(0, 100), H.history[\"loss\"], label=\"train_loss\")\n",
        "plt.plot(np.arange(0, 100), H.history[\"val_loss\"], label=\"val_loss\")\n",
        "plt.plot(np.arange(0, 100), H.history[\"accuracy\"], label=\"train_acc\")\n",
        "plt.plot(np.arange(0, 100), H.history[\"val_accuracy\"], label=\"val_acc\")\n",
        "plt.title(\"Training Loss and Accuracy\")\n",
        "plt.xlabel(\"Epoch\")\n",
        "plt.ylabel(\"Loss/Accuracy\")\n",
        "plt.legend()\n",
        "plt.show();"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 299
        },
        "id": "BNKKy65Vc-ul",
        "outputId": "c294c040-e243-43f0-9fc0-d86024785041"
      },
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "**NEURAL NETWORKS: CIFAR10**"
      ],
      "metadata": {
        "id": "sXQ6pUlJjxGe"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "#@ PREPARING THE DATASET:\n",
        "((trainX, trainY), (testX, testY)) = cifar10.load_data()            # Loading Dataset. \n",
        "trainX = trainX.astype(\"float\") / 255.0                             # Normalizing Dataset. \n",
        "testX = testX.astype(\"float\") / 255.0                               # Normalizing Dataset. \n",
        "trainX = trainX.reshape((trainX.shape[0], 3072))                    # Reshaping Dataset.\n",
        "testX = testX.reshape((testX.shape[0], 3072))                       # Reshaping Dataset. \n",
        "\n",
        "#@ PREPARING THE DATASET:\n",
        "lb = LabelBinarizer()                                               # Initializing LabelBinarizer. \n",
        "trainY = lb.fit_transform(trainY)                                   # Converting Labels to Vectors. \n",
        "testY = lb.transform(testY)                                         # Converting Labels to Vectors. \n",
        "labelNames = [\"airplane\", \"automobile\", \"bird\", \"cat\", \"deer\", \n",
        "              \"dog\", \"frog\", \"horse\", \"ship\", \"truck\"]              # Initializing LabelNames. "
      ],
      "metadata": {
        "id": "RO-tJWcrfRuo"
      },
      "execution_count": 18,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "#@ DEFINING NETWORK ARCHITECTURE: \n",
        "model = Sequential()                                                # Initializing Sequential Model. \n",
        "model.add(Dense(1024, input_shape=(3072, ), activation=\"relu\"))     # Adding Dense Layer.\n",
        "model.add(Dense(512, activation=\"relu\"))                            # Adding Dense Layer. \n",
        "model.add(Dense(10, activation=\"softmax\"))                          # Adding Softmax Layer. \n",
        "\n",
        "#@ TRAINING NEURAL NETWORK:\n",
        "sgd = SGD(0.01)                                                     # Initializing SGD Optimizer. \n",
        "model.compile(loss=\"categorical_crossentropy\", optimizer=sgd, \n",
        "              metrics=[\"accuracy\"])                                 # Compiling Model. \n",
        "H = model.fit(trainX, trainY,validation_data=(testX,testY), \n",
        "              epochs=100, batch_size=32)                            # Training Model. "
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "h6huPyOomG4G",
        "outputId": "9db05623-fec2-47d7-c133-514f5a8b4b5f"
      },
      "execution_count": 20,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 1.8406 - accuracy: 0.3420 - val_loss: 1.7348 - val_accuracy: 0.3825\n",
            "Epoch 2/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 1.6510 - accuracy: 0.4156 - val_loss: 1.6031 - val_accuracy: 0.4362\n",
            "Epoch 3/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.5671 - accuracy: 0.4468 - val_loss: 1.5518 - val_accuracy: 0.4554\n",
            "Epoch 4/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.5076 - accuracy: 0.4694 - val_loss: 1.4932 - val_accuracy: 0.4649\n",
            "Epoch 5/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.4636 - accuracy: 0.4840 - val_loss: 1.5371 - val_accuracy: 0.4583\n",
            "Epoch 6/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.4193 - accuracy: 0.5019 - val_loss: 1.5123 - val_accuracy: 0.4642\n",
            "Epoch 7/100\n",
            "1563/1563 [==============================] - 53s 34ms/step - loss: 1.3844 - accuracy: 0.5103 - val_loss: 1.4398 - val_accuracy: 0.4830\n",
            "Epoch 8/100\n",
            "1563/1563 [==============================] - 50s 32ms/step - loss: 1.3519 - accuracy: 0.5241 - val_loss: 1.4655 - val_accuracy: 0.4756\n",
            "Epoch 9/100\n",
            "1563/1563 [==============================] - 58s 37ms/step - loss: 1.3215 - accuracy: 0.5360 - val_loss: 1.4063 - val_accuracy: 0.5046\n",
            "Epoch 10/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.2932 - accuracy: 0.5451 - val_loss: 1.4873 - val_accuracy: 0.4698\n",
            "Epoch 11/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 1.2667 - accuracy: 0.5557 - val_loss: 1.4139 - val_accuracy: 0.5064\n",
            "Epoch 12/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.2412 - accuracy: 0.5635 - val_loss: 1.3832 - val_accuracy: 0.5135\n",
            "Epoch 13/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 1.2179 - accuracy: 0.5713 - val_loss: 1.3635 - val_accuracy: 0.5188\n",
            "Epoch 14/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 1.1932 - accuracy: 0.5820 - val_loss: 1.3539 - val_accuracy: 0.5180\n",
            "Epoch 15/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.1710 - accuracy: 0.5905 - val_loss: 1.4565 - val_accuracy: 0.4920\n",
            "Epoch 16/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.1484 - accuracy: 0.5978 - val_loss: 1.4600 - val_accuracy: 0.4899\n",
            "Epoch 17/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 1.1245 - accuracy: 0.6048 - val_loss: 1.3499 - val_accuracy: 0.5294\n",
            "Epoch 18/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.1036 - accuracy: 0.6138 - val_loss: 1.5200 - val_accuracy: 0.4766\n",
            "Epoch 19/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.0802 - accuracy: 0.6217 - val_loss: 1.3302 - val_accuracy: 0.5353\n",
            "Epoch 20/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 1.0575 - accuracy: 0.6295 - val_loss: 1.3436 - val_accuracy: 0.5337\n",
            "Epoch 21/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.0379 - accuracy: 0.6385 - val_loss: 1.3514 - val_accuracy: 0.5230\n",
            "Epoch 22/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 1.0156 - accuracy: 0.6453 - val_loss: 1.3569 - val_accuracy: 0.5272\n",
            "Epoch 23/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.9960 - accuracy: 0.6516 - val_loss: 1.3385 - val_accuracy: 0.5403\n",
            "Epoch 24/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.9717 - accuracy: 0.6627 - val_loss: 1.3174 - val_accuracy: 0.5410\n",
            "Epoch 25/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.9519 - accuracy: 0.6689 - val_loss: 1.3771 - val_accuracy: 0.5256\n",
            "Epoch 26/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.9288 - accuracy: 0.6766 - val_loss: 1.2957 - val_accuracy: 0.5525\n",
            "Epoch 27/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.9096 - accuracy: 0.6823 - val_loss: 1.3273 - val_accuracy: 0.5418\n",
            "Epoch 28/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.8909 - accuracy: 0.6912 - val_loss: 1.3505 - val_accuracy: 0.5308\n",
            "Epoch 29/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.8644 - accuracy: 0.7021 - val_loss: 1.4347 - val_accuracy: 0.5217\n",
            "Epoch 30/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.8480 - accuracy: 0.7034 - val_loss: 1.3321 - val_accuracy: 0.5516\n",
            "Epoch 31/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.8268 - accuracy: 0.7144 - val_loss: 1.3302 - val_accuracy: 0.5508\n",
            "Epoch 32/100\n",
            "1563/1563 [==============================] - 37s 24ms/step - loss: 0.8036 - accuracy: 0.7206 - val_loss: 1.4245 - val_accuracy: 0.5298\n",
            "Epoch 33/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.7827 - accuracy: 0.7289 - val_loss: 1.3971 - val_accuracy: 0.5496\n",
            "Epoch 34/100\n",
            "1563/1563 [==============================] - 37s 24ms/step - loss: 0.7637 - accuracy: 0.7349 - val_loss: 1.3286 - val_accuracy: 0.5496\n",
            "Epoch 35/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.7438 - accuracy: 0.7447 - val_loss: 1.4390 - val_accuracy: 0.5286\n",
            "Epoch 36/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.7228 - accuracy: 0.7518 - val_loss: 1.5199 - val_accuracy: 0.5196\n",
            "Epoch 37/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.7043 - accuracy: 0.7585 - val_loss: 1.3910 - val_accuracy: 0.5441\n",
            "Epoch 38/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.6861 - accuracy: 0.7634 - val_loss: 1.7789 - val_accuracy: 0.4579\n",
            "Epoch 39/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.6641 - accuracy: 0.7723 - val_loss: 1.4867 - val_accuracy: 0.5382\n",
            "Epoch 40/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.6405 - accuracy: 0.7811 - val_loss: 1.5240 - val_accuracy: 0.5259\n",
            "Epoch 41/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.6211 - accuracy: 0.7883 - val_loss: 1.5695 - val_accuracy: 0.5158\n",
            "Epoch 42/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.6029 - accuracy: 0.7939 - val_loss: 1.5064 - val_accuracy: 0.5307\n",
            "Epoch 43/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.5772 - accuracy: 0.8050 - val_loss: 1.5662 - val_accuracy: 0.5355\n",
            "Epoch 44/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.5624 - accuracy: 0.8079 - val_loss: 1.5629 - val_accuracy: 0.5326\n",
            "Epoch 45/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.5441 - accuracy: 0.8168 - val_loss: 1.4456 - val_accuracy: 0.5605\n",
            "Epoch 46/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.5298 - accuracy: 0.8214 - val_loss: 1.5117 - val_accuracy: 0.5464\n",
            "Epoch 47/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.5088 - accuracy: 0.8299 - val_loss: 1.6368 - val_accuracy: 0.5239\n",
            "Epoch 48/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.4889 - accuracy: 0.8373 - val_loss: 1.5817 - val_accuracy: 0.5371\n",
            "Epoch 49/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.4670 - accuracy: 0.8438 - val_loss: 1.5466 - val_accuracy: 0.5472\n",
            "Epoch 50/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.4531 - accuracy: 0.8507 - val_loss: 1.4838 - val_accuracy: 0.5615\n",
            "Epoch 51/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.4349 - accuracy: 0.8577 - val_loss: 1.5519 - val_accuracy: 0.5570\n",
            "Epoch 52/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.4203 - accuracy: 0.8613 - val_loss: 1.7107 - val_accuracy: 0.5356\n",
            "Epoch 53/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.4057 - accuracy: 0.8664 - val_loss: 1.5732 - val_accuracy: 0.5588\n",
            "Epoch 54/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.3873 - accuracy: 0.8723 - val_loss: 1.9268 - val_accuracy: 0.5003\n",
            "Epoch 55/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.3744 - accuracy: 0.8795 - val_loss: 1.8539 - val_accuracy: 0.5169\n",
            "Epoch 56/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.3512 - accuracy: 0.8881 - val_loss: 1.6008 - val_accuracy: 0.5615\n",
            "Epoch 57/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.3470 - accuracy: 0.8883 - val_loss: 1.6636 - val_accuracy: 0.5521\n",
            "Epoch 58/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.3288 - accuracy: 0.8947 - val_loss: 1.7198 - val_accuracy: 0.5433\n",
            "Epoch 59/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.3137 - accuracy: 0.9018 - val_loss: 1.7184 - val_accuracy: 0.5480\n",
            "Epoch 60/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.2980 - accuracy: 0.9062 - val_loss: 1.7242 - val_accuracy: 0.5563\n",
            "Epoch 61/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.2881 - accuracy: 0.9100 - val_loss: 1.7265 - val_accuracy: 0.5589\n",
            "Epoch 62/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.2763 - accuracy: 0.9132 - val_loss: 1.8206 - val_accuracy: 0.5452\n",
            "Epoch 63/100\n",
            "1563/1563 [==============================] - 40s 25ms/step - loss: 0.2597 - accuracy: 0.9203 - val_loss: 1.9619 - val_accuracy: 0.5275\n",
            "Epoch 64/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.2451 - accuracy: 0.9247 - val_loss: 1.7618 - val_accuracy: 0.5482\n",
            "Epoch 65/100\n",
            "1563/1563 [==============================] - 64s 41ms/step - loss: 0.2399 - accuracy: 0.9260 - val_loss: 1.9281 - val_accuracy: 0.5402\n",
            "Epoch 66/100\n",
            "1563/1563 [==============================] - 38s 24ms/step - loss: 0.2264 - accuracy: 0.9312 - val_loss: 1.7582 - val_accuracy: 0.5650\n",
            "Epoch 67/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.2107 - accuracy: 0.9375 - val_loss: 1.7887 - val_accuracy: 0.5646\n",
            "Epoch 68/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.2032 - accuracy: 0.9403 - val_loss: 1.8571 - val_accuracy: 0.5569\n",
            "Epoch 69/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1930 - accuracy: 0.9447 - val_loss: 1.8208 - val_accuracy: 0.5652\n",
            "Epoch 70/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1891 - accuracy: 0.9447 - val_loss: 2.0137 - val_accuracy: 0.5431\n",
            "Epoch 71/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1735 - accuracy: 0.9520 - val_loss: 1.8689 - val_accuracy: 0.5607\n",
            "Epoch 72/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.1651 - accuracy: 0.9537 - val_loss: 1.9140 - val_accuracy: 0.5605\n",
            "Epoch 73/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1514 - accuracy: 0.9583 - val_loss: 1.8836 - val_accuracy: 0.5653\n",
            "Epoch 74/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1458 - accuracy: 0.9611 - val_loss: 1.9325 - val_accuracy: 0.5586\n",
            "Epoch 75/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1364 - accuracy: 0.9640 - val_loss: 1.9432 - val_accuracy: 0.5595\n",
            "Epoch 76/100\n",
            "1563/1563 [==============================] - 40s 25ms/step - loss: 0.1311 - accuracy: 0.9662 - val_loss: 2.0359 - val_accuracy: 0.5524\n",
            "Epoch 77/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1207 - accuracy: 0.9696 - val_loss: 1.9438 - val_accuracy: 0.5654\n",
            "Epoch 78/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1183 - accuracy: 0.9696 - val_loss: 2.0794 - val_accuracy: 0.5517\n",
            "Epoch 79/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1060 - accuracy: 0.9743 - val_loss: 2.0091 - val_accuracy: 0.5677\n",
            "Epoch 80/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.1014 - accuracy: 0.9755 - val_loss: 2.0232 - val_accuracy: 0.5623\n",
            "Epoch 81/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0923 - accuracy: 0.9796 - val_loss: 2.1073 - val_accuracy: 0.5558\n",
            "Epoch 82/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0908 - accuracy: 0.9788 - val_loss: 2.0320 - val_accuracy: 0.5642\n",
            "Epoch 83/100\n",
            "1563/1563 [==============================] - 38s 25ms/step - loss: 0.0821 - accuracy: 0.9827 - val_loss: 2.4988 - val_accuracy: 0.5257\n",
            "Epoch 84/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0821 - accuracy: 0.9815 - val_loss: 2.0999 - val_accuracy: 0.5621\n",
            "Epoch 85/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0720 - accuracy: 0.9858 - val_loss: 2.3023 - val_accuracy: 0.5286\n",
            "Epoch 86/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0712 - accuracy: 0.9848 - val_loss: 2.1582 - val_accuracy: 0.5608\n",
            "Epoch 87/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0623 - accuracy: 0.9881 - val_loss: 2.1808 - val_accuracy: 0.5623\n",
            "Epoch 88/100\n",
            "1563/1563 [==============================] - 40s 25ms/step - loss: 0.0594 - accuracy: 0.9890 - val_loss: 2.1528 - val_accuracy: 0.5655\n",
            "Epoch 89/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0557 - accuracy: 0.9902 - val_loss: 2.1393 - val_accuracy: 0.5687\n",
            "Epoch 90/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0545 - accuracy: 0.9898 - val_loss: 2.1602 - val_accuracy: 0.5721\n",
            "Epoch 91/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0489 - accuracy: 0.9918 - val_loss: 2.2244 - val_accuracy: 0.5632\n",
            "Epoch 92/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0464 - accuracy: 0.9923 - val_loss: 2.1930 - val_accuracy: 0.5708\n",
            "Epoch 93/100\n",
            "1563/1563 [==============================] - 40s 26ms/step - loss: 0.0419 - accuracy: 0.9936 - val_loss: 2.2212 - val_accuracy: 0.5651\n",
            "Epoch 94/100\n",
            "1563/1563 [==============================] - 40s 25ms/step - loss: 0.0382 - accuracy: 0.9943 - val_loss: 2.2309 - val_accuracy: 0.5667\n",
            "Epoch 95/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0363 - accuracy: 0.9949 - val_loss: 2.2424 - val_accuracy: 0.5690\n",
            "Epoch 96/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0353 - accuracy: 0.9951 - val_loss: 2.2042 - val_accuracy: 0.5726\n",
            "Epoch 97/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0302 - accuracy: 0.9965 - val_loss: 2.2572 - val_accuracy: 0.5705\n",
            "Epoch 98/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0308 - accuracy: 0.9961 - val_loss: 2.2753 - val_accuracy: 0.5628\n",
            "Epoch 99/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0283 - accuracy: 0.9965 - val_loss: 2.3223 - val_accuracy: 0.5694\n",
            "Epoch 100/100\n",
            "1563/1563 [==============================] - 39s 25ms/step - loss: 0.0275 - accuracy: 0.9967 - val_loss: 2.2872 - val_accuracy: 0.5703\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#@ MODEL EVALUATION: \n",
        "predictions = model.predict(testX, batch_size=32)           # Getting Predictions. \n",
        "print(classification_report(testY.argmax(axis=1), \n",
        "                            predictions.argmax(axis=1),\n",
        "                            target_names=labelNames))       # Inspecting Classification Report. "
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "XtDt3eH8o4u0",
        "outputId": "f9a827b0-fcdd-4517-9003-870753593339"
      },
      "execution_count": 21,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "              precision    recall  f1-score   support\n",
            "\n",
            "    airplane       0.61      0.68      0.64      1000\n",
            "  automobile       0.72      0.63      0.67      1000\n",
            "        bird       0.46      0.51      0.48      1000\n",
            "         cat       0.40      0.39      0.39      1000\n",
            "        deer       0.55      0.42      0.48      1000\n",
            "         dog       0.47      0.47      0.47      1000\n",
            "        frog       0.62      0.63      0.63      1000\n",
            "       horse       0.64      0.62      0.63      1000\n",
            "        ship       0.65      0.73      0.69      1000\n",
            "       truck       0.60      0.62      0.61      1000\n",
            "\n",
            "    accuracy                           0.57     10000\n",
            "   macro avg       0.57      0.57      0.57     10000\n",
            "weighted avg       0.57      0.57      0.57     10000\n",
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#@ INSPECTING TRAINING LOSS AND ACCURACY:\n",
        "plt.style.use(\"ggplot\")\n",
        "plt.figure()\n",
        "plt.plot(np.arange(0, 100), H.history[\"loss\"], label=\"train_loss\")\n",
        "plt.plot(np.arange(0, 100), H.history[\"val_loss\"], label=\"val_loss\")\n",
        "plt.plot(np.arange(0, 100), H.history[\"accuracy\"], label=\"train_acc\")\n",
        "plt.plot(np.arange(0, 100), H.history[\"val_accuracy\"], label=\"val_acc\")\n",
        "plt.title(\"Training Loss and Accuracy\")\n",
        "plt.xlabel(\"Epoch\")\n",
        "plt.ylabel(\"Loss/Accuracy\")\n",
        "plt.legend()\n",
        "plt.show();"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 299
        },
        "id": "GGZpp1NSqhcp",
        "outputId": "5e3e94f9-cd70-4f4e-df5d-71e4eafe5f15"
      },
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {}
        }
      ]
    }
  ]
}