Convolutional auto encoder for MNIST

mnist_convautoencoder.py

import argparse
import torch
import torch.nn as nn
from torch.nn import Conv2d, ConvTranspose2d, Linear
import torch.nn.functional as F
import torch.optim as optim
from torchvision import transforms, datasets
from torchvision.utils import save_image
import os

class MNIST_CNN(nn.Module):
    def __init__(self):
        super(MNIST_CNN, self).__init__()
        self.enc_conv1 = Conv2d(1, 16, 3, 1, padding=1) # indim x indim x 16 volume
        self.enc_conv2 = Conv2d(16, 8, 3, 1, padding=1) # indim x indim x 16 volume
        self.enc_conv3 = Conv2d(8, 8, 3, 1, padding=1)
        self.dec_conv1 = ConvTranspose2d(8, 8, 5, 2, padding=1)
        self.dec_conv2 = ConvTranspose2d(8, 8, 2, 2, padding=0)
        self.dec_conv3 = ConvTranspose2d(8, 16, 2, 2, padding=0)
        self.logits = Conv2d(16, 1, 3, 1, padding=1)
    
    def encode(self, x):
        x = F.max_pool2d(F.relu(self.enc_conv1(x)), 2, 2)
        x = F.max_pool2d(F.relu(self.enc_conv2(x)), 2, 2)
        return F.max_pool2d(F.relu(self.enc_conv3(x)), 2, 2)

    def decode(self, x):
        x = F.relu(self.dec_conv1(x)) #F.interpolate(x, scale_factor=2, mode='nearest')))
        x = F.relu(self.dec_conv2(x)) #F.interpolate(x, scale_factor=2, mode='nearest')))
        return F.relu(self.dec_conv3(x)) #F.interpolate(x, scale_factor=2, mode='nearest')))
    
    def forward(self, x):
        x = self.encode(x)
        x = self.decode(x)
        return torch.sigmoid((self.logits(x)))

    
def train(args, model, device, train_loader, optimizer, epoch):
    model.train()
    train_loss = 0
    for batch_idx, (data, _) in enumerate(train_loader):
        data = data.to(device, dtype=torch.float64)
        optimizer.zero_grad()
        recon_data = model(data.float())
        if batch_idx and args.trace > 0: pdb.set_trace()
        loss = F.mse_loss(recon_data.float(), data.float(), reduction='sum')
        loss.backward()
        optimizer.step()
        if batch_idx % args.log_interval == 0:
            print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                epoch, batch_idx * len(data), len(train_loader.dataset), 100. * batch_idx / len(train_loader),
                loss.item() / len(data)))

def main():
    parser = argparse.ArgumentParser(description="Convolutional Autoencoder for MNIST")
    parser.add_argument('--batch-size', type=int, default=64, metavar='N',
                        help='input batch size for training (defaults to 64)')
    parser.add_argument('--test-batch-size', type=int, default=1000, metavar='N',
                        help='input batch size for testing (defaults to 1000)')
    parser.add_argument('--epochs', type=int, default=200, metavar='N',
                        help='number of epochs to train (defaults to 10)')
    parser.add_argument('--no-cuda', action='store_true', default=False, 
                        help='disables CUDA training')
    parser.add_argument('--trace', action='store_true', default=False, 
                        help='turns on a pdb trace token')
    parser.add_argument('--lr', type=float, default=0.001, metavar='LR',
                        help='learning rate (defaults to 0.01)')
    parser.add_argument('--momentum', type=float, default=0.5, metavar='M',
                        help='SGD momentum (defaults to 0.5)')
    parser.add_argument('--seed', type=int, default=1, metavar='S',
                        help='random seed (defaults to 1)')
    parser.add_argument('--log-interval', type=int, default=10, metavar='N',
                        help='how many batches to wait before logging training status')
    args = parser.parse_args()
    torch.manual_seed(args.seed)
    use_cuda = not args.no_cuda and torch.cuda.is_available()
    print("Using CUDA" if use_cuda else "Not using CUDA")
    device = torch.device("cuda" if use_cuda else "cpu")
    kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}
    train_loader = torch.utils.data.DataLoader(datasets.MNIST('./MNIST', train=True, download=True, 
                                                                    transform=transforms.Compose([transforms.ToTensor(), 
                                                                                                  transforms.Normalize((0.1307,), (0.3081,))])), 
                                                     batch_size=args.batch_size, shuffle=True, **kwargs)

    model = MNIST_CNN().to(device)
    optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum)

    for epoch in range(1, args.epochs + 1):
        train(args, model, device, train_loader, optimizer, epoch)

if __name__ == "__main__":
    main()