如何在 PyTorch 中更新神经网络的参数? [英] How can I update the parameters of a neural network in PyTorch?

问题描述

假设我想在 PyTorch(继承自 torch.nn.Module) 通过 0.9.我该怎么做?

Let's say I wanted to multiply all parameters of a neural network in PyTorch (an instance of a class inheriting from torch.nn.Module) by 0.9. How would I do that?

推荐答案

让 net 成为你的神经网络类的一个实例.然后你可以做

Let net an instance of your neural network class. You can then do

state_dict = net.state_dict()

for name, param in state_dict.items():
    # Transform the parameter as required.
    transformed_param = param * 0.9

    # Update the parameter.
    state_dict[name].copy_(transformed_param)

将所有参数乘以0.9.

如果你只想更新权重而不是所有参数，你可以这样做

If you ever only want to update weights instead of all parameters, you can do

state_dict = net.state_dict()

for name, param in state_dict.items():
    # Don't update if this is not a weight.
    if not "weight" in name:
        continue

    # Transform the parameter as required.
    transformed_param = param * 0.9

    # Update the parameter.
    state_dict[name].copy_(transformed_param)

这篇关于如何在 PyTorch 中更新神经网络的参数?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！