pytorch获取模型某一层参数名及参数值方式

Z时代
2024-01-10
分类：综合

1、Motivation：

I wanna modify the value of some param;

I wanna check the value of some param.

The needed function:

2、state_dict() #generator type

model.modules()#generator type

named_parameters()#OrderDict type


from torch import nn
import torch
#creat a simple model
model = nn.Sequential(
  nn.Conv3d(1,16,kernel_size=1),
  nn.Conv3d(16,2,kernel_size=1))#tend to print the W of this layer
input = torch.randn([1,1,16,256,256])
if torch.cuda.is_available():
  print('cuda is avaliable')
  model.cuda()
  input = input.cuda()
#打印某一层的参数名
for name in model.state_dict():
  print(name)
#Then I konw that the name of target layer is '1.weight'
#schemem1(recommended)
print(model.state_dict()['1.weight'])
#scheme2
params = list(model.named_parameters())#get the index by debuging
print(params[2][0])#name
print(params[2][1].data)#data
#scheme3
params = {}#change the tpye of 'generator' into dict
for name,param in model.named_parameters():
params[name] = param.detach().cpu().numpy()
print(params['0.weight'])
#scheme4
for layer in model.modules():
if(isinstance(layer,nn.Conv3d)):
  print(layer.weight)
#打印每一层的参数名和参数值
#schemem1(recommended)
for name,param in model.named_parameters():
  print(name，param)
#scheme2
for name in model.state_dict():
  print(name)
  print(model.state_dict()[name])