pytorch中的Parameter函數(shù)可以對某個張量進(jìn)行參數(shù)化。它可以將不可訓(xùn)練的張量轉(zhuǎn)化為可訓(xùn)練的參數(shù)類型，同時將轉(zhuǎn)化后的張量綁定到模型可訓(xùn)練參數(shù)的列表中，當(dāng)更新模型的參數(shù)時一并將其更新。

torch.nn.parameter.Parameter

data (Tensor)：表示需要參數(shù)化的張量
requires_grad (bool, optional)：表示是否該張量是否需要梯度，默認(rèn)值為True

代碼介紹

pytorch中的Parameter函數(shù)具體的代碼示例如下所示

import torch
import torch.nn as nn
class NeuralNetwork(nn.Module):
	def __init__(self, input_dim, output_dim):
		super(NeuralNetwork, self).__init__()
		self.linear = nn.Linear(input_dim, output_dim)
		self.linear.weight = torch.nn.Parameter(torch.zeros(input_dim, output_dim))
		self.linear.bias = torch.nn.Parameter(torch.ones(output_dim))
	def forward(self, input_array):
		output = self.linear(input_array)
		return output
if __name__ == '__main__':
	net = NeuralNetwork(4, 6)
	for param in net.parameters():
		print(param)

代碼的結(jié)果如下所示：

當(dāng)神經(jīng)網(wǎng)絡(luò)的參數(shù)不是用Parameter函數(shù)參數(shù)化直接賦值給權(quán)重參數(shù)時，則會報錯，具體的程序

import torch
import torch.nn as nn
class NeuralNetwork(nn.Module):
	def __init__(self, input_dim, output_dim):
		super(NeuralNetwork, self).__init__()
		self.linear = nn.Linear(input_dim, output_dim)
		self.linear.weight = torch.zeros(input_dim, output_dim)
		self.linear.bias = torch.ones(output_dim)
	def forward(self, input_array):
		output = self.linear(input_array)
		return output
if __name__ == '__main__':
	net = NeuralNetwork(4, 6)
	for param in net.parameters():
		print(param)

代碼運行報錯結(jié)果如下所示: