初始化火炬中的权重和偏差

Question

Pytorch 中以下 keras 代码的等效命令是什么

Dense(64, kernel_initializer='he_normal', bias_initializer='zeros', name='uhat_digitcaps')(d5)

如何初始化权重和偏差？

谢谢！

Answer 1

class Net(nn.Module):
    def __init__(self, in_channels, out_channels):
        self.linear = nn.Linear(in_channels, 64)
        nn.init.kaiming_normal_(self.linear.weight, mode='fan_out')
        nn.init.constant_(self.linear.bias, 0)

初始化火炬中的权重和偏差

Initialize weights and bias in torch

torch

keras

pytorch