Variational_AutoEncoders.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "gpuType": "T4",
      "authorship_tag": "ABX9TyMYUuVLdAeVdMixJFJJeXPe",
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/Ashu-00/AutoEncodersAndGenerators/blob/main/Variational_AutoEncoders.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "6UJ8gnZSA45N"
      },
      "outputs": [],
      "source": [
        "import torch\n",
        "import torchvision\n",
        "import torchvision.transforms as transforms\n",
        "import torch.nn as nn\n",
        "import torch.nn.functional as F\n",
        "import torch.optim as optim\n",
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "from tqdm import tqdm"
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "class VAE(nn.Module):\n",
        "  def __init__(self,insize=28*28, latent_size=200):\n",
        "    super().__init__()\n",
        "\n",
        "    self.h=latent_size\n",
        "    self.ins=insize\n",
        "\n",
        "    self.encoder = nn.Sequential(\n",
        "        nn.Linear(self.ins,512),\n",
        "        nn.LeakyReLU(),\n",
        "        nn.Linear(512,400),\n",
        "        nn.LeakyReLU(),\n",
        "\n",
        "    )\n",
        "\n",
        "    self.mumake=nn.Linear(400, self.h)\n",
        "    self.varmake=nn.Linear(400, self.h)\n",
        "\n",
        "    self.decoder= nn.Sequential(\n",
        "        nn.Linear(self.h,512),\n",
        "        nn.LeakyReLU(),\n",
        "        nn.Linear(512,self.ins),\n",
        "        nn.Sigmoid(),\n",
        "\n",
        "    )\n",
        "\n",
        "  def reparameterize(self, mu, var):\n",
        "\n",
        "    eps = torch.randn_like(var)\n",
        "    return mu + eps * var\n",
        "\n",
        "  def forward(self, x):\n",
        "    x = torch.flatten(x, start_dim=1)\n",
        "\n",
        "    x=self.encoder(x)\n",
        "\n",
        "    mu=self.mumake(x)\n",
        "    lvar=self.varmake(x)\n",
        "\n",
        "    z=self.reparameterize(mu, lvar)\n",
        "\n",
        "\n",
        "    x=self.decoder(z)\n",
        "    return x.view(-1,1,28,28), mu, lvar"
      ],
      "metadata": {
        "id": "p_iJz_lsBWM5"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "x=torch.randn(1,1,28,28)\n",
        "model=VAE()\n",
        "x, mu, lvar=model(x)\n",
        "print(x.shape)\n",
        "print(mu.shape)\n",
        "print(lvar.shape)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "bGd_GrnyGtZv",
        "outputId": "ce2921f8-737a-44ab-f756-4126374a1ffa"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "torch.Size([1, 1, 28, 28])\n",
            "torch.Size([1, 200])\n",
            "torch.Size([1, 200])\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "from torchvision import datasets\n",
        "from torch.utils.data import DataLoader"
      ],
      "metadata": {
        "id": "wi74Ynm7Lpcb"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "dataset=datasets.MNIST(root='./data', train=True, transform=transforms.ToTensor(), download=True)"
      ],
      "metadata": {
        "id": "sKhriayfMLmv"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "generator1 = torch.Generator().manual_seed(42)\n",
        "trainset,testset=torch.utils.data.random_split(dataset,[.9,.1],generator=generator1)"
      ],
      "metadata": {
        "id": "mNzu935OMfm6"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "trainloader=torch.utils.data.DataLoader(trainset,batch_size=32,shuffle=True)\n",
        "testloader=torch.utils.data.DataLoader(testset,batch_size=32,shuffle=False)\n",
        "device=torch.device(\"cuda\" if torch.cuda.is_available() else 'cpu')\n"
      ],
      "metadata": {
        "id": "M3CWzB2PN524"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def lossfunc(realx, outx, mu,lvar):\n",
        "    reproduction_loss = nn.functional.binary_cross_entropy(outx,realx, reduction='sum')\n",
        "    KLD = - 0.5 * torch.sum(1+ torch.log(lvar.pow(2)) - mu.pow(2) - lvar.pow(2))\n",
        "\n",
        "    return reproduction_loss + KLD"
      ],
      "metadata": {
        "id": "1CjrpaxBODuy"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "def train(model,train_loader,epochs=3,lr=9e-3,device=\"cpu\"):\n",
        "  model.train()\n",
        "  model.to(device)\n",
        "  optimizer=optim.Adam(model.parameters(),lr=lr)\n",
        "\n",
        "  for epoch in range(epochs):\n",
        "    loop=tqdm(enumerate((train_loader)))\n",
        "    lossavg=0\n",
        "    for (bid,(inp,lab)) in loop:\n",
        "      inp=inp.to(device)\n",
        "\n",
        "      #lab=lab.to(device)\n",
        "      out,mu,sigma=model(inp)\n",
        "      loss=lossfunc(inp,out,mu,sigma)\n",
        "\n",
        "      optimizer.zero_grad()\n",
        "      loss.backward()\n",
        "      optimizer.step()\n",
        "      loop.set_postfix(loss=loss.item())\n",
        "\n",
        "      lossavg+=loss.item()\n",
        "\n",
        "    print(f\"epoch:{epoch} loss:{lossavg/(len(train_loader))}\")\n"
      ],
      "metadata": {
        "id": "n1DUf0UOUSm0"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "\n",
        "model=VAE().to(device)"
      ],
      "metadata": {
        "id": "pbjehCVmZRPo"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model.load_state_dict(torch.load('model.pt'))"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "c_NobJ48jScK",
        "outputId": "9599048e-37c4-4ea9-b5d0-785db781a673"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<All keys matched successfully>"
            ]
          },
          "metadata": {},
          "execution_count": 109
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "#model=VAE().to(device)\n",
        "train(model,trainloader,epochs=10,lr=5e-4,device=device)"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "78NF3YZMVV6N",
        "outputId": "ef6a0f83-9f17-492b-dd57-03e1a94c85ea"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:17, 96.64it/s, loss=1.52e+3] \n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:0 loss:3355.6900009950755\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:18, 92.63it/s, loss=1.49e+3]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:1 loss:3354.8387636302205\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:17, 96.34it/s, loss=1.76e+3] \n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:2 loss:3351.750536588696\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:17, 95.13it/s, loss=1.6e+3]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:3 loss:3349.458480473378\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:18, 93.68it/s, loss=1.6e+3] \n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:4 loss:3344.959348994974\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:17, 95.46it/s, loss=1.7e+3]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:5 loss:3343.4360541031824\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:18, 92.05it/s, loss=1.49e+3] \n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:6 loss:3339.2012825192996\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:17, 95.44it/s, loss=1.77e+3]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:7 loss:3335.197641092454\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:18, 92.68it/s, loss=1.75e+3]\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:8 loss:3332.9043787373184\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "1688it [00:17, 95.55it/s, loss=1.82e+3]"
          ]
        },
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "epoch:9 loss:3330.577905880896\n"
          ]
        },
        {
          "output_type": "stream",
          "name": "stderr",
          "text": [
            "\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "torch.save(model.state_dict(),'model.pt')"
      ],
      "metadata": {
        "id": "GJ2rCDDUZWkl"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "model.to(\"cpu\")\n",
        "img=testset[197][0]\n",
        "out,mu,sigma=model(img.unsqueeze(0))\n",
        "plt.subplot(1,2,1)\n",
        "plt.imshow(img.detach().cpu().squeeze(0,1),cmap='gray')\n",
        "plt.subplot(1,2,2)\n",
        "plt.imshow(out.detach().cpu().squeeze(0,1),cmap='gray')\n",
        "plt.show()"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 287
        },
        "id": "3-pmCN6ojiqN",
        "outputId": "1267704d-366a-485d-90cf-2a6e1d186af5"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 2 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    },
    {
      "cell_type": "code",
      "source": [
        "def inference(dataset):\n",
        "    \"\"\"\n",
        "    Generates (num_examples) of a particular digit.\n",
        "    Specifically we extract an example of each digit,\n",
        "    then after we have the mu, sigma representation for\n",
        "    each digit we can sample from that.\n",
        "\n",
        "    After we sample we can run the decoder part of the VAE\n",
        "    and generate examples.\n",
        "    \"\"\"\n",
        "    images = []\n",
        "    idx = 0\n",
        "    for x, y in dataset:\n",
        "        if y == idx:\n",
        "            images.append(x)\n",
        "            idx += 1\n",
        "        if idx == 10:\n",
        "            break\n",
        "\n",
        "    encodings_digit = []\n",
        "    for d in range(10):\n",
        "        with torch.no_grad():\n",
        "            x = model.encoder(images[d].view(1, 784))\n",
        "            mu=model.mumake(x)\n",
        "            sigma=model.varmake(x)\n",
        "\n",
        "\n",
        "        encodings_digit.append((mu, sigma))\n",
        "\n",
        "\n",
        "    for i,(mu, sigma) in enumerate(encodings_digit):\n",
        "        epsilon = torch.randn_like(5*sigma)\n",
        "        z = mu + sigma * epsilon\n",
        "        out = model.decoder(z)\n",
        "        out = out.view(-1, 1, 28, 28)\n",
        "        plt.subplot(1,10,i+1)\n",
        "        plt.title(f\"{i}\")\n",
        "        plt.imshow(out.detach().cpu().squeeze(0,1),cmap='gray')\n",
        "        plt.axis('off')\n",
        "    plt.show()\n",
        "\n",
        "model.to(\"cpu\")\n",
        "inference(trainset)\n"
      ],
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 101
        },
        "id": "NuGC_Gu6lApr",
        "outputId": "f6752d11-8e3d-437a-e8d4-05224b50a241"
      },
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 640x480 with 10 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {}
        }
      ]
    }
  ]
}