Classifying movie reviews (IMDB dataset) : Binary Classification.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "authorship_tag": "ABX9TyMiVA8oy2I15ajLANw2kbP4",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/kalaiselvan-t/Deep-Learning/blob/master/Classifying%20movie%20reviews%20(IMDB%20dataset)%20%3A%20Binary%20Classification.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Classifying Movie Reviews from the IMDB dataset"
      ],
      "metadata": {
        "id": "4HUUMXIxqXZT"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "opkSxLmikgj1"
      },
      "outputs": [],
      "source": [
        "'''load the dataset'''\n",
        "\n",
        "from keras.datasets import imdb\n",
        "\n",
        "(train_data, train_labels), (test_data, test_labels) = imdb.load_data(num_words=10000) "
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "'''\n",
        "- Know what words correspond to each index\n",
        "- To visualize the dataset\n",
        "'''\n",
        "\n",
        "# train_labels[0]\n",
        "word_index = imdb.get_word_index()"
      ],
      "metadata": {
        "id": "0bqpn44slVPi"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Decode Reviews back to English"
      ],
      "metadata": {
        "id": "j08lp1OUnsD6"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "'''\n",
        "- Decode the indexes back to english words\n",
        "'''\n",
        "\n",
        "reverse_word_index = dict([(value,key) for (key,value) in word_index.items()])\n",
        "\n",
        "decoded_review = ' '.join([reverse_word_index.get(i-3,'?') for i in train_data[0]])\n",
        "\n",
        "# decoded_review"
      ],
      "metadata": {
        "id": "x8SUe8uhl44i"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Preparing the data"
      ],
      "metadata": {
        "id": "1aFo25l6n5AI"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "import numpy as np\n",
        "\n",
        "# Function to vectorize the input data\n",
        "def vectorize_sequences(sequences, dimensions=10000):\n",
        "  results = np.zeros((len(sequences),dimensions))\n",
        "  for i,sequence in enumerate(sequences):\n",
        "    results[i,sequence] = 1\n",
        "  return results"
      ],
      "metadata": {
        "id": "uhdu77--n61_"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "'''\n",
        "- Preprocessing the data\n",
        "- Split the data into training and test sets\n",
        "'''\n",
        "x_train = vectorize_sequences(train_data)\n",
        "x_test = vectorize_sequences(test_data)\n",
        "print('x_train shape: ', x_train.shape)\n",
        "print('x_test shape: ', x_test.shape)\n",
        "\n",
        "y_train = np.asarray(train_labels).astype('float32')\n",
        "y_test = np.asarray(test_labels).astype('float32')\n",
        "print('y_train shape: ', y_train.shape)\n",
        "print('y_test shape: ', y_test.shape)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "M3c_7-ZspFBt",
        "outputId": "2d519cf4-fa46-4d0e-b14d-4dbe8c429edf"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "x_train shape:  (25000, 10000)\n",
            "x_test shape:  (25000, 10000)\n",
            "y_train shape:  (25000,)\n",
            "y_test shape:  (25000,)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "x_train[0]\n",
        "# y_train"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "aZGuHkhNp1Ab",
        "outputId": "e2b2fe25-9943-46ec-d9ff-2374c46653e7"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([0., 1., 1., ..., 0., 0., 0.])"
            ]
          },
          "metadata": {},
          "execution_count": 8
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Building the network - Overfitting Demostration"
      ],
      "metadata": {
        "id": "vSgLeKNfqhzM"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "'''\n",
        "- Demonstration of overfitting\n",
        "- Look for peaks in the validation loss around 4 epochs\n",
        "- That means overfitting\n",
        "'''\n",
        "\n",
        "from keras import models\n",
        "from keras import layers\n",
        "\n",
        "model = models.Sequential()\n",
        "model.add(layers.Dense(16, activation='relu', input_shape=(10000,)))\n",
        "model.add(layers.Dense(16, activation='relu'))\n",
        "model.add(layers.Dense(1, activation='sigmoid'))"
      ],
      "metadata": {
        "id": "sQNGXJFOql94"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model.compile(optimizer='rmsprop', \n",
        "              loss='binary_crossentropy',\n",
        "              metrics=['acc'])"
      ],
      "metadata": {
        "id": "VmVjn5v9tHTf"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Setting aside validation data"
      ],
      "metadata": {
        "id": "0Zw6aPPjt0JP"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "'''\n",
        "- Set aside validation data to monitor the model accuracy\n",
        "'''\n",
        "x_val = x_train[:10000]\n",
        "partial_x_train = x_train[10000:]\n",
        "\n",
        "y_val = y_train[:10000]\n",
        "partial_y_train = y_train[10000:]"
      ],
      "metadata": {
        "id": "E2aKfaTxtzV4"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Training the model"
      ],
      "metadata": {
        "id": "Na-PQlqfuwN5"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "history = model.fit(partial_x_train,\n",
        "                    partial_y_train,\n",
        "                    epochs=20,\n",
        "                    batch_size=512,\n",
        "                    validation_data=(x_val,y_val))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "WHDh-rCnuyIg",
        "outputId": "7eac3783-7943-4c6b-d2a4-467172de1059"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/20\n",
            "30/30 [==============================] - 3s 69ms/step - loss: 0.5362 - acc: 0.7879 - val_loss: 0.4201 - val_acc: 0.8549\n",
            "Epoch 2/20\n",
            "30/30 [==============================] - 1s 40ms/step - loss: 0.3441 - acc: 0.8891 - val_loss: 0.3238 - val_acc: 0.8814\n",
            "Epoch 3/20\n",
            "30/30 [==============================] - 1s 43ms/step - loss: 0.2576 - acc: 0.9151 - val_loss: 0.2914 - val_acc: 0.8861\n",
            "Epoch 4/20\n",
            "30/30 [==============================] - 2s 69ms/step - loss: 0.2109 - acc: 0.9287 - val_loss: 0.2878 - val_acc: 0.8836\n",
            "Epoch 5/20\n",
            "30/30 [==============================] - 1s 45ms/step - loss: 0.1763 - acc: 0.9417 - val_loss: 0.2772 - val_acc: 0.8875\n",
            "Epoch 6/20\n",
            "30/30 [==============================] - 1s 40ms/step - loss: 0.1527 - acc: 0.9495 - val_loss: 0.2915 - val_acc: 0.8835\n",
            "Epoch 7/20\n",
            "30/30 [==============================] - 1s 40ms/step - loss: 0.1305 - acc: 0.9575 - val_loss: 0.2945 - val_acc: 0.8820\n",
            "Epoch 8/20\n",
            "30/30 [==============================] - 1s 40ms/step - loss: 0.1133 - acc: 0.9659 - val_loss: 0.3246 - val_acc: 0.8804\n",
            "Epoch 9/20\n",
            "30/30 [==============================] - 1s 39ms/step - loss: 0.0995 - acc: 0.9693 - val_loss: 0.3408 - val_acc: 0.8738\n",
            "Epoch 10/20\n",
            "30/30 [==============================] - 1s 41ms/step - loss: 0.0856 - acc: 0.9755 - val_loss: 0.3304 - val_acc: 0.8811\n",
            "Epoch 11/20\n",
            "30/30 [==============================] - 1s 40ms/step - loss: 0.0729 - acc: 0.9807 - val_loss: 0.3847 - val_acc: 0.8747\n",
            "Epoch 12/20\n",
            "30/30 [==============================] - 1s 39ms/step - loss: 0.0676 - acc: 0.9822 - val_loss: 0.3673 - val_acc: 0.8789\n",
            "Epoch 13/20\n",
            "30/30 [==============================] - 1s 47ms/step - loss: 0.0536 - acc: 0.9879 - val_loss: 0.3876 - val_acc: 0.8779\n",
            "Epoch 14/20\n",
            "30/30 [==============================] - 2s 57ms/step - loss: 0.0491 - acc: 0.9881 - val_loss: 0.4069 - val_acc: 0.8774\n",
            "Epoch 15/20\n",
            "30/30 [==============================] - 1s 47ms/step - loss: 0.0415 - acc: 0.9913 - val_loss: 0.4303 - val_acc: 0.8749\n",
            "Epoch 16/20\n",
            "30/30 [==============================] - 1s 38ms/step - loss: 0.0356 - acc: 0.9930 - val_loss: 0.4562 - val_acc: 0.8743\n",
            "Epoch 17/20\n",
            "30/30 [==============================] - 1s 41ms/step - loss: 0.0312 - acc: 0.9942 - val_loss: 0.4728 - val_acc: 0.8710\n",
            "Epoch 18/20\n",
            "30/30 [==============================] - 1s 39ms/step - loss: 0.0261 - acc: 0.9957 - val_loss: 0.4996 - val_acc: 0.8734\n",
            "Epoch 19/20\n",
            "30/30 [==============================] - 1s 40ms/step - loss: 0.0206 - acc: 0.9974 - val_loss: 0.5195 - val_acc: 0.8687\n",
            "Epoch 20/20\n",
            "30/30 [==============================] - 1s 40ms/step - loss: 0.0187 - acc: 0.9975 - val_loss: 0.5419 - val_acc: 0.8696\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "# Visualizing the loss and accuracy\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "\n",
        "history_dict = history.history\n",
        "loss_values = history_dict['loss']\n",
        "val_loss_values = history_dict['val_loss']\n",
        "acc_values = history_dict['acc']\n",
        "val_acc_values = history_dict['val_acc']\n",
        "\n",
        "epochs = range(1,len(acc_values)+1)\n",
        "\n",
        "plt.plot(epochs, loss_values, 'bo', label='Training loss')\n",
        "plt.plot(epochs, val_loss_values, 'b', label=\"Validation loss\")\n",
        "plt.title('Training and Validation loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "plt.show()\n",
        "\n",
        "plt.clf()\n",
        "plt.plot(epochs, acc_values, 'bo', label=\"Training acc\")\n",
        "plt.plot(epochs, val_acc_values, 'b' , label=\"Validation acc\")\n",
        "plt.title('Training and Validation accuracy')\n",
        "plt.xlabel('Epochs')\n",
        "plt.ylabel('Loss')\n",
        "plt.legend()\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 573
        },
        "id": "rEdrGDUZvlSO",
        "outputId": "080575af-0bf0-4a61-94a8-a62d7b55e6d5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "##Building the Network - Mitigating Overfitting"
      ],
      "metadata": {
        "id": "pCOpOrkxyyYm"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "'''\n",
        "- New model to mitigate the overfitting\n",
        "- Reduce the number of epochs to 4\n",
        "'''\n",
        "\n",
        "from keras.engine import input_spec\n",
        "model = models.Sequential()\n",
        "model.add(layers.Dense(16, activation='relu', input_shape=(10000,)))\n",
        "model.add(layers.Dense(16, activation='relu'))\n",
        "model.add(layers.Dense(1, activation='sigmoid'))\n",
        "\n",
        "model.compile(optimizer='rmsprop',\n",
        "              loss='binary_crossentropy',\n",
        "              metrics=['accuracy'])\n",
        "\n",
        "history = model.fit(x_train, y_train, epochs=4, batch_size=512)\n",
        "\n",
        "results = model.evaluate(x_test, y_test)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "-RlRgyacy55y",
        "outputId": "9e829c49-39a8-423a-f70c-e52183c5e7b3"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/4\n",
            "49/49 [==============================] - 5s 80ms/step - loss: 0.4483 - accuracy: 0.8164\n",
            "Epoch 2/4\n",
            "49/49 [==============================] - 1s 30ms/step - loss: 0.2664 - accuracy: 0.9057\n",
            "Epoch 3/4\n",
            "49/49 [==============================] - 1s 30ms/step - loss: 0.2071 - accuracy: 0.9268\n",
            "Epoch 4/4\n",
            "49/49 [==============================] - 1s 29ms/step - loss: 0.1790 - accuracy: 0.9359\n",
            "782/782 [==============================] - 2s 3ms/step - loss: 0.3473 - accuracy: 0.8631\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "print(results)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2N6KUaxZ0hCX",
        "outputId": "2f22601d-4604-459f-da49-4f07cc40d005"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "[0.3472737967967987, 0.8630800247192383]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "##Predicting Review on new data with trained model"
      ],
      "metadata": {
        "id": "2nX0fwWr0t8u"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "model.predict(x_test)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "2I6Fjs9J04eg",
        "outputId": "614d960e-ebcb-4052-eca5-41fb5a56ef4c"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "782/782 [==============================] - 3s 3ms/step\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([[0.3029611 ],\n",
              "       [0.99985576],\n",
              "       [0.97630805],\n",
              "       ...,\n",
              "       [0.17468607],\n",
              "       [0.16229758],\n",
              "       [0.8885039 ]], dtype=float32)"
            ]
          },
          "metadata": {},
          "execution_count": 17
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "##Bridge\n",
        "\n",
        "#### Future Experiments\n",
        "\n",
        "1. Modify the number of hidden layers\n",
        "2. Modify the number of hidden units\n",
        "3. Use different loss functions\n",
        "4. Use different activation functions\n",
        "\n",
        "**Goal:** To increase the accuracy from 88% to 95%(state-of-the-art) or more"
      ],
      "metadata": {
        "id": "j6DZrLnf1NRP"
      }
    }
  ]
}