PyTorch, nn.Sequential(), access weights of a specific module in nn.Sequential()

Question

this should be a quick one. When I use a pre-defined module in PyTorch, I can typically access its weights fairly easily. However, how do I access them if I wrapped the modu

Answer 1

From the PyTorch forum, this is the recommended way:

model_2.layer[0].weight

Answer 2

You can access modules by name using _modules:

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()

        self.conv1 = nn.Conv2d(3, 3, 3)

    def forward(self, input):
        return self.conv1(input)

model = Net()
print(model._modules['conv1'])

Answer 3

An easy way to access the weights is to use the state_dict() of your model.

This should work in your case:

for k, v in model_2.state_dict().iteritems():
    print("Layer {}".format(k))
    print(v)

Another option is to get the modules() iterator. If you know beforehand the type of your layers this should also work:

for layer in model_2.modules():
   if isinstance(layer, nn.Linear):
        print(layer.weight)