33.pdf

    1. import  torch
    2. import  torch.nn as nn
    3. import  torch.nn.functional as F
    4. import  torch.optim as optim
    5. from    torchvision import datasets, transforms
    7. from visdom import Visdom
    9. batch_size=200
    10. learning_rate=0.01
    11. epochs=10
    13. train_loader = torch.utils.data.DataLoader(
    14.     datasets.MNIST('../data', train=True, download=True,
    15.                    transform=transforms.Compose([
    16.                        transforms.ToTensor(),
    17.                        # transforms.Normalize((0.1307,), (0.3081,))
    18.                    ])),
    19.     batch_size=batch_size, shuffle=True)
    20. test_loader = torch.utils.data.DataLoader(
    21.     datasets.MNIST('../data', train=False, transform=transforms.Compose([
    22.         transforms.ToTensor(),
    23.         # transforms.Normalize((0.1307,), (0.3081,))
    24.     ])),
    25.     batch_size=batch_size, shuffle=True)
    29. class MLP(nn.Module):
    31.     def __init__(self):
    32.         super(MLP, self).__init__()
    34.         self.model = nn.Sequential(
    35.             nn.Linear(784, 200),
    36.             nn.LeakyReLU(inplace=True),
    37.             nn.Linear(200, 200),
    38.             nn.LeakyReLU(inplace=True),
    39.             nn.Linear(200, 10),
    40.             nn.LeakyReLU(inplace=True),
    41.         )
    43.     def forward(self, x):
    44.         x = self.model(x)
    46.         return x
    48. device = torch.device('cuda:0')
    49. net = MLP().to(device)
    50. optimizer = optim.SGD(net.parameters(), lr=learning_rate, weight_decay=0.01)
    51. criteon = nn.CrossEntropyLoss().to(device)
    53. viz = Visdom()
    55. viz.line([0.], [0.], win='train_loss', opts=dict(title='train loss'))
    56. viz.line([[0.0, 0.0]], [0.], win='test', opts=dict(title='test loss&acc.',
    57.                                                    legend=['loss', 'acc.']))
    58. global_step = 0
    60. for epoch in range(epochs):
    62.     for batch_idx, (data, target) in enumerate(train_loader):
    63.         data = data.view(-1, 28*28)
    64.         data, target = data.to(device), target.cuda()
    66.         logits = net(data)
    67.         loss = criteon(logits, target)
    69.         optimizer.zero_grad()
    70.         loss.backward()
    71.         # print(w1.grad.norm(), w2.grad.norm())
    72.         optimizer.step()
    74.         global_step += 1
    75.         viz.line([loss.item()], [global_step], win='train_loss', update='append')
    77.         if batch_idx % 100 == 0:
    78.             print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
    79.                 epoch, batch_idx * len(data), len(train_loader.dataset),
    80.                        100. * batch_idx / len(train_loader), loss.item()))
    83.     test_loss = 0
    84.     correct = 0
    85.     for data, target in test_loader:
    86.         data = data.view(-1, 28 * 28)
    87.         data, target = data.to(device), target.cuda()
    88.         logits = net(data)
    89.         test_loss += criteon(logits, target).item()
    91.         pred = logits.argmax(dim=1)
    92.         correct += pred.eq(target).float().sum().item()
    94.     viz.line([[test_loss, correct / len(test_loader.dataset)]],
    95.              [global_step], win='test', update='append')
    96.     viz.images(data.view(-1, 1, 28, 28), win='x')
    97.     viz.text(str(pred.detach().cpu().numpy()), win='pred',
    98.              opts=dict(title='pred'))
    100.     test_loss /= len(test_loader.dataset)
    101.     print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
    102.         test_loss, correct, len(test_loader.dataset),
    103.         100. * correct / len(test_loader.dataset)))