Multi Label Classification in pytorch

45

Here is a very simple dummy example:

model = nn.Linear(20, 5) # predict logits for 5 classes
x = torch.randn(1, 20)
y = torch.tensor([[1., 0., 1., 0., 0.]]) # get classA and classC as active

criterion = nn.BCEWithLogitsLoss()
optimizer = optim.SGD(model.parameters(), lr=1e-1)

for epoch in range(20):
    optimizer.zero_grad()
    output = model(x)
    loss = criterion(output, y)
    loss.backward()
    optimizer.step()
    print('Loss: {:.3f}'.format(loss.item()))
38 Likes