In PyTorch how are layer weights and biases initialized by default?

Question

I was wondering how are layer weights and biases initialized by default? E.g. if I create the linear layer torch.nn.Linear(5,100) How are weights and biases for this layer initialized by default?

Answer 1

PyTorch 1.0

Most layers are initialized using Kaiming Uniform method. Example layers include Linear, Conv2d, RNN etc. If you are using other layers, you should look up that layer on this doc. If it says weights are initialized using U(...) then its Kaiming Uniform method. Bias is initialized using LeCunn init, i.e., uniform(-std, std) where standard deviation std is 1/sqrt(fan_in) (code).

PyTorch 0.4.1, 0.3.1

Weights and biases are initialized using LeCunn init (see sec 4.6) for conv layers (code: 0.3.1, 0.4.1).

If you want to override default initialization then see this answer.

Answer 2

The weights are initilized in the following way:

def reset_parameters(self):
    stdv = 1. / math.sqrt(self.weight.size(1))
    self.weight.data.uniform_(-stdv, stdv)
    if self.bias is not None:
        self.bias.data.uniform_(-stdv, stdv)

https://github.com/pytorch/pytorch/blob/master/torch/nn/modules/linear.py#L48-L52