26.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
    8. batch_size=200
    9. learning_rate=0.01
    10. epochs=10
    12. train_loader = torch.utils.data.DataLoader(
    13.     datasets.MNIST('../data', train=True, download=True,
    14.                    transform=transforms.Compose([
    15.                        transforms.ToTensor(),
    16.                        transforms.Normalize((0.1307,), (0.3081,))
    17.                    ])),
    18.     batch_size=batch_size, shuffle=True)
    19. test_loader = torch.utils.data.DataLoader(
    20.     datasets.MNIST('../data', train=False, transform=transforms.Compose([
    21.         transforms.ToTensor(),
    22.         transforms.Normalize((0.1307,), (0.3081,))
    23.     ])),
    24.     batch_size=batch_size, shuffle=True)
    28. w1, b1 = torch.randn(200, 784, requires_grad=True),\
    29.          torch.zeros(200, requires_grad=True)
    30. w2, b2 = torch.randn(200, 200, requires_grad=True),\
    31.          torch.zeros(200, requires_grad=True)
    32. w3, b3 = torch.randn(10, 200, requires_grad=True),\
    33.          torch.zeros(10, requires_grad=True)
    35. torch.nn.init.kaiming_normal_(w1)
    36. torch.nn.init.kaiming_normal_(w2)
    37. torch.nn.init.kaiming_normal_(w3)
    40. def forward(x):
    41.     x = x@w1.t() + b1
    42.     x = F.relu(x)
    43.     x = x@w2.t() + b2
    44.     x = F.relu(x)
    45.     x = x@w3.t() + b3
    46.     x = F.relu(x)
    47.     return x
    51. optimizer = optim.SGD([w1, b1, w2, b2, w3, b3], lr=learning_rate)
    52. criteon = nn.CrossEntropyLoss()
    54. for epoch in range(epochs):
    56.     for batch_idx, (data, target) in enumerate(train_loader):
    57.         data = data.view(-1, 28*28)
    59.         logits = forward(data)
    60.         loss = criteon(logits, target)
    62.         optimizer.zero_grad()
    63.         loss.backward()
    64.         # print(w1.grad.norm(), w2.grad.norm())
    65.         optimizer.step()
    67.         if batch_idx % 100 == 0:
    68.             print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
    69.                 epoch, batch_idx * len(data), len(train_loader.dataset),
    70.                        100. * batch_idx / len(train_loader), loss.item()))
    73.     test_loss = 0
    74.     correct = 0
    75.     for data, target in test_loader:
    76.         data = data.view(-1, 28 * 28)
    77.         logits = forward(data)
    78.         test_loss += criteon(logits, target).item()
    80.         pred = logits.data.max(1)[1]
    81.         correct += pred.eq(target.data).sum()
    83.     test_loss /= len(test_loader.dataset)
    84.     print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(
    85.         test_loss, correct, len(test_loader.dataset),
    86.         100. * correct / len(test_loader.dataset)))