Lab_6/GenAI_Lab6_Autoencoder.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "execution_count": 1,
      "metadata": {
        "id": "HTt_JyWBYEk4"
      },
      "outputs": [],
      "source": [
        "import tensorflow as tf\n",
        "from tensorflow.keras.datasets import mnist\n",
        "from tensorflow.keras.models import Model\n",
        "from tensorflow.keras.layers import Input, Dense"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "(x_train, _), (x_test, _) = mnist.load_data()"
      ],
      "metadata": {
        "id": "470rezMHYH8d",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "11732f50-6dcf-46b8-b208-8999625e4628"
      },
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Downloading data from https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz\n",
            "11490434/11490434 [==============================] - 0s 0us/step\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "x_train = x_train.astype('float32') / 255.\n",
        "x_test = x_test.astype('float32') / 255."
      ],
      "metadata": {
        "id": "PAgHySCIYLYe"
      },
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x_train = x_train.reshape((len(x_train), -1))\n",
        "x_test = x_test.reshape((len(x_test), -1))"
      ],
      "metadata": {
        "id": "Lb_h6acdYM1h"
      },
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "encoding_dim = 32"
      ],
      "metadata": {
        "id": "pD2HIQLdYO1-"
      },
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "input_img = Input(shape=(784,))"
      ],
      "metadata": {
        "id": "V7qTYl7vYQYd"
      },
      "execution_count": 6,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "encoded = Dense(128, activation='relu')(input_img)\n",
        "encoded = Dense(64, activation='relu')(encoded)\n",
        "encoded = Dense(encoding_dim, activation='relu')(encoded)"
      ],
      "metadata": {
        "id": "WSQO5j8-YSK9"
      },
      "execution_count": 7,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "decoded = Dense(64, activation='relu')(encoded)\n",
        "decoded = Dense(128, activation='relu')(decoded)\n",
        "decoded = Dense(784, activation='sigmoid')(decoded)"
      ],
      "metadata": {
        "id": "5UI3aeHjYUZ1"
      },
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "autoencoder = Model(input_img, decoded)"
      ],
      "metadata": {
        "id": "uVM6dcexYWHl"
      },
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "encoder = Model(input_img, encoded)"
      ],
      "metadata": {
        "id": "tSDAvJHcYXd5"
      },
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "encoded_input = Input(shape=(encoding_dim,))\n",
        "decoder_layer1 = autoencoder.layers[-3]\n",
        "decoder_layer2 = autoencoder.layers[-2]\n",
        "decoder_layer3 = autoencoder.layers[-1]\n",
        "decoder = Model(encoded_input, decoder_layer3(decoder_layer2(decoder_layer1(encoded_input))))"
      ],
      "metadata": {
        "id": "VYyiaAtyYYqU"
      },
      "execution_count": 16,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "autoencoder.compile(optimizer='adam', loss='binary_crossentropy')"
      ],
      "metadata": {
        "id": "mYvKlW1yYbF0"
      },
      "execution_count": 17,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "autoencoder.fit(x_train, x_train, epochs=50, batch_size=256,shuffle=True, validation_data=(x_test, x_test))"
      ],
      "metadata": {
        "id": "yQnvUNRYYdNp",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "ec53184f-4b7e-479a-85e2-8dab2a667eaa"
      },
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/50\n",
            "235/235 [==============================] - 6s 19ms/step - loss: 0.2453 - val_loss: 0.1657\n",
            "Epoch 2/50\n",
            "235/235 [==============================] - 5s 20ms/step - loss: 0.1499 - val_loss: 0.1372\n",
            "Epoch 3/50\n",
            "235/235 [==============================] - 4s 17ms/step - loss: 0.1332 - val_loss: 0.1253\n",
            "Epoch 4/50\n",
            "235/235 [==============================] - 4s 18ms/step - loss: 0.1226 - val_loss: 0.1172\n",
            "Epoch 5/50\n",
            "235/235 [==============================] - 5s 20ms/step - loss: 0.1161 - val_loss: 0.1126\n",
            "Epoch 6/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.1124 - val_loss: 0.1093\n",
            "Epoch 7/50\n",
            "235/235 [==============================] - 4s 18ms/step - loss: 0.1094 - val_loss: 0.1072\n",
            "Epoch 8/50\n",
            "235/235 [==============================] - 5s 20ms/step - loss: 0.1071 - val_loss: 0.1050\n",
            "Epoch 9/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.1051 - val_loss: 0.1029\n",
            "Epoch 10/50\n",
            "235/235 [==============================] - 4s 17ms/step - loss: 0.1035 - val_loss: 0.1018\n",
            "Epoch 11/50\n",
            "235/235 [==============================] - 5s 20ms/step - loss: 0.1020 - val_loss: 0.1001\n",
            "Epoch 12/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.1006 - val_loss: 0.0987\n",
            "Epoch 13/50\n",
            "235/235 [==============================] - 4s 18ms/step - loss: 0.0994 - val_loss: 0.0979\n",
            "Epoch 14/50\n",
            "235/235 [==============================] - 5s 20ms/step - loss: 0.0984 - val_loss: 0.0970\n",
            "Epoch 15/50\n",
            "235/235 [==============================] - 4s 17ms/step - loss: 0.0974 - val_loss: 0.0963\n",
            "Epoch 16/50\n",
            "235/235 [==============================] - 4s 18ms/step - loss: 0.0967 - val_loss: 0.0952\n",
            "Epoch 17/50\n",
            "235/235 [==============================] - 5s 20ms/step - loss: 0.0958 - val_loss: 0.0948\n",
            "Epoch 18/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0950 - val_loss: 0.0936\n",
            "Epoch 19/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0942 - val_loss: 0.0932\n",
            "Epoch 20/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0934 - val_loss: 0.0922\n",
            "Epoch 21/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0929 - val_loss: 0.0923\n",
            "Epoch 22/50\n",
            "235/235 [==============================] - 4s 17ms/step - loss: 0.0924 - val_loss: 0.0917\n",
            "Epoch 23/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0919 - val_loss: 0.0909\n",
            "Epoch 24/50\n",
            "235/235 [==============================] - 4s 17ms/step - loss: 0.0916 - val_loss: 0.0906\n",
            "Epoch 25/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0912 - val_loss: 0.0902\n",
            "Epoch 26/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0908 - val_loss: 0.0897\n",
            "Epoch 27/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0904 - val_loss: 0.0896\n",
            "Epoch 28/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0900 - val_loss: 0.0891\n",
            "Epoch 29/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0896 - val_loss: 0.0889\n",
            "Epoch 30/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0893 - val_loss: 0.0884\n",
            "Epoch 31/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0889 - val_loss: 0.0880\n",
            "Epoch 32/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0886 - val_loss: 0.0879\n",
            "Epoch 33/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0883 - val_loss: 0.0876\n",
            "Epoch 34/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0880 - val_loss: 0.0877\n",
            "Epoch 35/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0878 - val_loss: 0.0869\n",
            "Epoch 36/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0875 - val_loss: 0.0868\n",
            "Epoch 37/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0872 - val_loss: 0.0865\n",
            "Epoch 38/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0870 - val_loss: 0.0863\n",
            "Epoch 39/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0869 - val_loss: 0.0864\n",
            "Epoch 40/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0867 - val_loss: 0.0861\n",
            "Epoch 41/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0865 - val_loss: 0.0858\n",
            "Epoch 42/50\n",
            "235/235 [==============================] - 4s 17ms/step - loss: 0.0863 - val_loss: 0.0861\n",
            "Epoch 43/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0863 - val_loss: 0.0857\n",
            "Epoch 44/50\n",
            "235/235 [==============================] - 5s 22ms/step - loss: 0.0860 - val_loss: 0.0859\n",
            "Epoch 45/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0859 - val_loss: 0.0854\n",
            "Epoch 46/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0858 - val_loss: 0.0853\n",
            "Epoch 47/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0857 - val_loss: 0.0856\n",
            "Epoch 48/50\n",
            "235/235 [==============================] - 4s 16ms/step - loss: 0.0855 - val_loss: 0.0853\n",
            "Epoch 49/50\n",
            "235/235 [==============================] - 4s 17ms/step - loss: 0.0854 - val_loss: 0.0853\n",
            "Epoch 50/50\n",
            "235/235 [==============================] - 5s 21ms/step - loss: 0.0853 - val_loss: 0.0849\n"
          ]
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.src.callbacks.History at 0x7e7c5e28cdf0>"
            ]
          },
          "metadata": {},
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt\n",
        "# Use the encoder to transform the test set into the encoded space\n",
        "encoded_imgs = encoder.predict(x_test)\n",
        "# Use the decoder to generate new images from the encoded representations\n",
        "generated_images = decoder.predict(encoded_imgs)\n",
        "# Reshape and visualize the generated images\n",
        "plt.figure(figsize=(20, 4))\n",
        "n=10\n",
        "for i in range(n):\n",
        "  ax = plt.subplot(2, n, i + 1)\n",
        "  plt.imshow(generated_images[i].reshape(28, 28))\n",
        "  plt.gray()\n",
        "  ax.get_xaxis().set_visible(False)\n",
        "  ax.get_yaxis().set_visible(False)\n",
        "plt.show()\n"
      ],
      "metadata": {
        "id": "aN6bTAYpYnil",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 114
        },
        "outputId": "76d3a4dd-5e0e-4ff8-f615-e6e2659efb96"
      },
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "313/313 [==============================] - 1s 2ms/step\n",
            "313/313 [==============================] - 1s 2ms/step\n"
          ]
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 2000x400 with 10 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "In the above implementation, the autoencoder is trained on the MNIST dataset, which consists of handwritten digits from 0 to 9. The autoencoder learns to encode and decode representations of these digits during the training process. Therefore, the generated images produced by the decoder will likely resemble digits from the MNIST dataset."
      ],
      "metadata": {
        "id": "QDEkyRiGecDc"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "###Observations:\n",
        "\n",
        "1. The generated images likely resembles the digits present in the training dataset because the model has learned to encode and decode these specific patterns.\n",
        "\n",
        "2. Not Completely New Images: Since the model is trained on a specific dataset, it may not be able to generate entirely new images that it has never seen before.\n",
        "\n",
        " The autoencoder is constrained by the patterns present in the MNIST dataset, and it can only generate variations or reconstructions of those patterns.\n",
        "\n",
        "\n",
        "---\n",
        "\n"
      ],
      "metadata": {
        "id": "SKiapdx_eh_9"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "**Variational Autoencoders** (VAEs) explicitly model the distribution of the encoded representations and allows us to sample from this distribution to generate new images.\n",
        "\n",
        "In a VAE, the encoder produces not only the mean but also the variance of the distribution for each dimension in the encoded space. This means that rather than providing a fixed point in the encoded space, VAEs model a probability distribution. Sampling from this distribution during the generation process allows the model to generate diverse and novel outputs.\n",
        "\n",
        "\n",
        "- Latent Space Distribution: VAEs model the latent space as a probability distribution, typically a Gaussian distribution (normal distribution), where the mean and variance are learned during training.\n",
        "\n",
        "- Reparameterization Trick: To enable backpropagation during training, a reparameterization trick is used.\n",
        "\n",
        " Instead of directly sampling from the distribution, the model samples from a standard Gaussian distribution and then scales and shifts the samples using the mean and variance predicted by the encoder.\n",
        "\n",
        "- Sampling for Generation: During the generation process, we can sample from the learned distribution in the latent space to obtain different representations. This allows the generation of diverse and novel images by exploring different regions of the latent space."
      ],
      "metadata": {
        "id": "WjZ6I0VwfECH"
      }
    }
  ]
}