Skip to content

Commit

Permalink
Add GAN HW
Browse files Browse the repository at this point in the history
  • Loading branch information
@airogachev
airogachev committed Dec 2, 2022
1 parent 612b5dd commit 4bdc736
Showing 1 changed file with 319 additions and 0 deletions.
319 changes: 319 additions & 0 deletions 13-GANs/GAN_HW.ipynb
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,319 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<a href=\"https://colab.research.google.com/github/HSE-LAMBDA/MLDM-2022/blob/main/13-GANs/GAN_homework.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### WGAN\n",
"\n",
"* Modify snippets below and implement [Wasserstein GAN](https://arxiv.org/abs/1701.07875) ([From GAN to WGAN\n",
"](https://lilianweng.github.io/posts/2017-08-20-gan/)) with weight clipping. (2 points)\n",
"\n",
"* Replace weight clipping with [gradient penalty](https://arxiv.org/pdf/1704.00028v3.pdf). (2 points)\n",
"\n",
"* Add labels into WGAN, performing [conditional generation](https://arxiv.org/pdf/1411.1784.pdf). (2 points) \n",
"\n",
"Write a report about experiments and results, add plots and visualizations."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import torch\n",
"import torch.nn as nn\n",
"import torch.nn.functional as F\n",
"import torch.optim as optim\n",
"from torch.utils.data import DataLoader, Dataset\n",
"\n",
"import torchvision\n",
"import matplotlib.pyplot as plt\n",
"import numpy as np\n",
"\n",
"from torch.autograd import Variable"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Creating config object (argparse workaround)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"class Config:\n",
" pass\n",
"\n",
"config = Config()\n",
"config.mnist_path = None\n",
"config.batch_size = 16\n",
"config.num_workers = 3\n",
"config.num_epochs = 10\n",
"config.noise_size = 50\n",
"config.print_freq = 100\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Create dataloder"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"train = torchvision.datasets.FashionMNIST(\"fashion_mnist\", train=True, transform=torchvision.transforms.ToTensor(), download=True)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"dataloader = DataLoader(train, batch_size=16, shuffle=True)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"len(dataloader)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"for image, cat in dataloader:\n",
" break"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"scrolled": true
},
"outputs": [],
"source": [
"image.size()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Create generator and discriminator"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"class Generator(nn.Module):\n",
" def __init__(self):\n",
" super(Generator, self).__init__()\n",
" self.model = nn.Sequential( \n",
" nn.Linear(config.noise_size, 200),\n",
" nn.ReLU(inplace=True),\n",
" nn.Linear(200, 28*28),\n",
" nn.Sigmoid())\n",
" \n",
" def forward(self, x):\n",
" return self.model(x)\n",
" \n",
"class Discriminator(nn.Module):\n",
" def __init__(self):\n",
" super(Discriminator, self).__init__()\n",
" self.model = nn.Sequential(\n",
" nn.Linear(28*28, 200),\n",
" nn.ReLU(inplace=True),\n",
" nn.Linear(200, 50),\n",
" nn.ReLU(inplace=True),\n",
" nn.Linear(50, 1), \n",
" nn.Sigmoid())\n",
" def forward(self, x):\n",
" return self.model(x)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"generator = Generator()\n",
"discriminator = Discriminator()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Create optimizers and loss"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"optim_G = optim.Adam(params=generator.parameters(), lr=0.0001)\n",
"optim_D = optim.Adam(params=discriminator.parameters(), lr=0.0001)\n",
"\n",
"criterion = nn.BCELoss()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Create necessary variables"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"input = Variable(torch.FloatTensor(config.batch_size, 28*28))\n",
"noise = Variable(torch.FloatTensor(config.batch_size, config.noise_size))\n",
"fixed_noise = Variable(torch.FloatTensor(config.batch_size, config.noise_size).normal_(0, 1))\n",
"label = Variable(torch.FloatTensor(config.batch_size))\n",
"real_label = 1\n",
"fake_label = 0"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### GAN"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"scrolled": true
},
"outputs": [],
"source": [
"ERRD_x = np.zeros(config.num_epochs)\n",
"ERRD_z = np.zeros(config.num_epochs)\n",
"ERRG = np.zeros(config.num_epochs)\n",
"N = len(dataloader)\n",
"\n",
"for epoch in range(config.num_epochs):\n",
" for iteration, (images, cat) in enumerate(dataloader):\n",
" ####### \n",
" # Discriminator stage: maximize log(D(x)) + log(1 - D(G(z))) \n",
" #######\n",
" discriminator.zero_grad()\n",
" \n",
" # real\n",
" label.data.fill_(real_label)\n",
" input_data = images.view(images.shape[0], -1)\n",
" output = discriminator(input_data)\n",
" errD_x = criterion(output, label)\n",
" ERRD_x[epoch] += errD_x.item()\n",
" errD_x.backward()\n",
" \n",
" # fake \n",
" noise.data.normal_(0, 1)\n",
" fake = generator(noise)\n",
" label.data.fill_(fake_label)\n",
" output = discriminator(fake.detach())\n",
" errD_z = criterion(output, label)\n",
" ERRD_z[epoch] += errD_z.item()\n",
" errD_z.backward()\n",
" \n",
" optim_D.step()\n",
" \n",
" ####### \n",
" # Generator stage: maximize log(D(G(x))\n",
" #######\n",
" generator.zero_grad()\n",
" label.data.fill_(real_label)\n",
" output = discriminator(fake)\n",
" errG = criterion(output, label)\n",
" ERRG[epoch] += errG.item()\n",
" errG.backward()\n",
" \n",
" optim_G.step()\n",
" \n",
" if (iteration+1) % config.print_freq == 0:\n",
" print('Epoch:{} Iter: {} errD_x: {:.2f} errD_z: {:.2f} errG: {:.2f}'.format(epoch+1,\n",
" iteration+1, \n",
" errD_x.item(),\n",
" errD_z.item(), \n",
" errG.item()))"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"noise.data.normal_(0, 1)\n",
"fake = generator(noise)\n",
"\n",
"plt.figure(figsize=(6, 7))\n",
"for i in range(16):\n",
" plt.subplot(4, 4, i + 1)\n",
" plt.imshow(fake[i].detach().numpy().reshape(28, 28), cmap=plt.cm.Greys_r)\n",
" plt.axis('off')"
]
}
],
"metadata": {
"anaconda-cloud": {},
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.8.11"
}
},
"nbformat": 4,
"nbformat_minor": 1
}

0 comments on commit 4bdc736

Please sign in to comment.