PyTorch: Extract learned weights correctly

问题:

I am trying to extract the weights from a linear layer, but they do not appear to change, although error is dropping monotonously (i.e. training is happening). Printing the weights' sum, nothing happens because it stays constant:

np.sum(model.fc2.weight.data.numpy())

Here are the code snippets:

def train(epochs):
    model.train()
    for epoch in range(1, epochs+1):
        # Train on train set
        print(np.sum(model.fc2.weight.data.numpy()))
        for batch_idx, (data, target) in enumerate(train_loader):
            data, target = Variable(data), Variable(data)
            optimizer.zero_grad()
            output = model(data)
            loss = criterion(output, target)
            loss.backward()
            optimizer.step()

and

# Define model
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        # an affine operation: y = Wx + b
        self.fc1 = nn.Linear(100, 80, bias=False)
        init.normal(self.fc1.weight, mean=0, std=1)
        self.fc2 = nn.Linear(80, 87)
        self.fc3 = nn.Linear(87, 94)
        self.fc4 = nn.Linear(94, 100)

    def forward(self, x):
        x = self.fc1(x)
        x = F.relu(self.fc2(x))
        x = F.relu(self.fc3(x))
        x = F.relu(self.fc4(x))
        return x

Maybe I am looking on the wrong parameters, although I checked the docs. Thanks for your help!

回答1:

Use model.parameters() to get trainable weight for any model or layer. Remember to put it inside list(), or you cannot print it out.

The following code snip worked

>>> import torch
>>> import torch.nn as nn
>>> l = nn.Linear(3,5)
>>> w = list(l.parameters())
>>> w