在用pdb debug的时候,有时候需要看一下特定layer的权重以及相应的梯度信息,如何查看呢?
1. 首先把你的模型打印出来,像这样
2. 然后观察到model下面有module的key,module下面有features的key, features下面有(0)的key,这样就可以直接打印出weight了,在pdb debug界面输入p model.module.features[0].weight,就可以看到weight,输入 p model.module.features[0].weight.grad就可以查看梯度信息
补充知识:查看Pytorch网络的各层输出(feature map)、权重(weight)、偏置(bias)
BatchNorm2d参数量
1
2
3
|
torch.nn.BatchNorm2d(num_features, eps = 1e - 05 , momentum = 0.1 , affine = True , track_running_stats = True )
# 卷积层中卷积核的数量C
num_features – C from an expected input of size (N, C, H, W)
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
|
>>> import torch
>>> m = torch.nn.BatchNorm2d( 100 )
>>> m.weight.shape
torch.Size([ 100 ])
>>> m.numel()
AttributeError: 'BatchNorm2d' object has no attribute 'numel'
>>> m.weight.numel()
100
>>> m.parameters().numel()
Traceback (most recent call last):
File "<stdin>" , line 1 , in <module>
AttributeError: 'generator' object has no attribute 'numel'
>>> [p.numel() for p in m.parameters()]
[ 100 , 100 ]
|
linear层
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
|
>>> import torch
>>> m1 = torch.nn.Linear( 100 , 10 )
# 参数数量= (输入神经元+1)*输出神经元
>>> m1.weight.shape
torch.Size([ 10 , 100 ])
>>> m1.bias.shape
torch.Size([ 10 ])
>>> m1.bias.numel()
10
>>> m1.weight.numel()
1000
>>> m11 = list (m1.parameters())
>>> m11[ 0 ].shape
# weight
torch.Size([ 10 , 100 ])
>>> m11[ 1 ].shape
# bias
torch.Size([ 10 ])
|
weight and bias
1
2
3
4
5
6
7
|
# Method 1 查看Parameters的方式多样化,直接访问即可
model = alexnet(pretrained = True ).to(device)
conv1_weight = model.features[ 0 ].weight # Method 2
# 这种方式还适合你想自己参考一个预训练模型写一个网络,各层的参数不变,但网络结构上表述有所不同
# 这样你就可以把param迭代出来,赋给你的网络对应层,避免直接load不能匹配的问题!
for layer,param in model.state_dict().items(): # param is weight or bias(Tensor)
print layer,param
|
feature map
由于pytorch是动态网络,不存储计算数据,查看各层输出的特征图并不是很方便!分下面两种情况讨论:
1、你想查看的层是独立的,那么你在forward时用变量接收并返回即可!!
1
2
3
4
5
6
7
8
9
|
class Net(nn.Module):
def __init__( self ):
self .conv1 = nn.Conv2d( 1 , 1 , 3 )
self .conv2 = nn.Conv2d( 1 , 1 , 3 )
self .conv3 = nn.Conv2d( 1 , 1 , 3 ) def forward( self , x):
out1 = F.relu( self .conv1(x))
out2 = F.relu( self .conv2(out1))
out3 = F.relu( self .conv3(out2))
return out1, out2, out3
|
2、你的想看的层在nn.Sequential()顺序容器中,这个麻烦些,主要有以下几种思路:
1
2
3
4
5
6
7
8
9
10
|
# Method 1 巧用nn.Module.children()
# 在模型实例化之后,利用nn.Module.children()删除你查看的那层的后面层
import torch
import torch.nn as nn
from torchvision import modelsmodel = models.alexnet(pretrained = True ) # remove last fully-connected layer
new_classifier = nn.Sequential( * list (model.classifier.children())[: - 1 ])
model.classifier = new_classifier
# Third convolutional layer
new_features = nn.Sequential( * list (model.features.children())[: 5 ])
model.features = new_features
|
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
|
# Method 2 巧用hook,推荐使用这种方式,不用改变原有模型
# torch.nn.Module.register_forward_hook(hook)
# hook(module, input, output) -> Nonemodel = models.alexnet(pretrained=True)
# 定义
def hook (module, input ,output):
print output.size()
# 注册
handle = model.features[ 0 ].register_forward_hook(hook)
# 删除句柄
handle.remove() # torch.nn.Module.register_backward_hook(hook)
# hook(module, grad_input, grad_output) -> Tensor or None
model = alexnet(pretrained = True ).to(device)
outputs = []
def hook (module, input ,output):
outputs.append(output)
print len (outputs)handle = model.features[ 0 ].register_backward_hook(hook)
|
注:还可以通过定义一个提取特征的类,甚至是重构成各层独立相同模型将问题转化成第一种
计算模型参数数量
def count_parameters(model):
return sum(p.numel() for p in model.parameters() if p.requires_grad)
以上这篇pytorch查看模型weight与grad方式就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。
原文链接:https://www.cnblogs.com/yongjieShi/p/10337174.html