PyTorch中模型的parameters()方法浅析

首先先定义一个模型:

import torch as t
import torch.nn as nn

class A(nn.Module):
    
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(2, 2, 3)
        self.conv2 = nn.Conv2d(2, 2, 3)
        self.conv3 = nn.Conv2d(2, 2, 3)
        
    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.conv3(x)
        
        return x

然后打印出该模型的参数：

pythona = A()
print(a.parameters())　#

以上代码说明parameters()会返回一个生成器（迭代器）

然后将其迭代打印出来：

print(list(a.parameters()))：#将迭代器转换成列表

Parameter containing:
tensor([[[[-0.0299,  0.0891,  0.0303],
          [ 0.0869, -0.0230, -0.1760],
          [ 0.1408,  0.0348,  0.1795]],

         [[ 0.2001,  0.0023, -0.1775],
          [ 0.0947, -0.0231, -0.1756],
          [ 0.1201, -0.0997, -0.0303]]],


        [[[-0.0425,  0.0748, -0.1754],
          [-0.1191, -0.1203, -0.1219],
          [-0.0794,  0.0895, -0.1719]],

         [[ 0.1968, -0.0463,  0.0550],
          [-0.0386,  0.1594,  0.1282],
          [-0.0009,  0.2167, -0.1783]]]], requires_grad=True)
Parameter containing:
tensor([ 0.0147, -0.0406], requires_grad=True)
Parameter containing:
tensor([[[[-0.0578, -0.1114, -0.1194],
          [-0.1469, -0.1175, -0.1616],
          [-0.2289, -0.0975, -0.1700]],

         [[-0.0894,  0.0074,  0.1222],
          [-0.0176, -0.0509,  0.1622],
          [-0.0405, -0.1349,  0.1782]]],


        [[[-0.0739,  0.2167,  0.1864],
          [ 0.0956, -0.1761,  0.0464],
          [ 0.0062, -0.0685,  0.0748]],

         [[ 0.1085,  0.1481,  0.1334],
          [ 0.2236, -0.0706, -0.0224],
          [ 0.0079, -0.1835, -0.0407]]]], requires_grad=True)
Parameter containing:
tensor([-8.0720e-05,  1.6026e-01], requires_grad=True)
Parameter containing:
tensor([[[[-0.0702,  0.1846,  0.0419],
          [-0.1891, -0.0893, -0.0024],
          [-0.0349, -0.0213,  0.0936]],

         [[-0.1062,  0.1242,  0.0391],
          [-0.1924,  0.0535, -0.1480],
          [ 0.0400, -0.0487, -0.2317]]],


        [[[ 0.1202,  0.0961,  0.2336],
          [ 0.2225, -0.2294, -0.2283],
          [-0.0963, -0.0311, -0.2354]],

         [[ 0.0676, -0.0439, -0.0962],
          [-0.2316, -0.0639, -0.0671],
          [ 0.1737, -0.1169, -0.1751]]]], requires_grad=True)
Parameter containing:
tensor([-0.1939, -0.0959], requires_grad=True)

从以上结果可以看出列表中有6个元素，由于nn.Conv2d()的参数包括self.weight和self.bias两部分，所以每个2D卷积层包括两部分的参数．注意self.bias是加在每个通道上的，所以self.bias的长度与output_channl相同

总结：

parameters()会返回一个生成器（迭代器），生成器每次生成的是Tensor类型的数据．