How to get all weights of RNN in PyTorch

1

Suppose I have one hidden layer fc, with layer size 30, input size 5, output size 2.

I can get weight between hidden layer and output layer by fc.weight.grad. But how could I get the weight that between input layer and fc?

2

Your code will get you the gradient of the fc.weight parameter, not the weights themselves.
If you would like to get the parameter values directly, you should call fc.weight.
I guess your input feature size should be 5 and the hidden dimension 30.
Here is a small code snippet printing the weights of both layers for this model:

in_features = 5
hidden = 30
out_features = 2

model = nn.Sequential(
    nn.Linear(in_features, hidden),
    nn.ReLU(),
    nn.Linear(hidden, out_features)
)

x = torch.randn(1, in_features)
output = model(x)
print(model[0].weight)
print(model[2].weight)
1 Like