获取Pytorch中间某一层权重或者特征

问题:训练好的网络模型想知道中间某一层的权重或者看看中间某一层的特征,如何处理呢?

1.获取某一层权重,并保存到excel中;

以resnet18为例说明:

import torch
import pandas as pd
import numpy as np
import torchvision.models as models

resnet18 = models.resnet18(pretrained=True)

parm={}
for name,parameters in resnet18.named_parameters():
    print(name,':',parameters.size())
    parm[name]=parameters.detach().numpy()

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9
• 10
• 11
• 12

​ 上述代码将每个模块参数存入parm字典中,parameters.detach().numpy()将tensor类型变量转换成numpy array形式,方便后续存储到表格中.输出为:

conv1.weight : torch.Size([64, 3, 7, 7])
bn1.weight : torch.Size([64])
bn1.bias : torch.Size([64])
layer1.0.conv1.weight : torch.Size([64, 64, 3, 3])
layer1.0.bn1.weight : torch.Size([64])
layer1.0.bn1.bias : torch.Size([64])
layer1.0.conv2.weight : torch.Size([64, 64, 3, 3])
layer1.0.bn2.weight : torch.Size([64])
layer1.0.bn2.bias : torch.Size([64])
layer1.1.conv1.weight : torch.Size([64, 64, 3, 3])
layer1.1.bn1.weight : torch.Size([64])
layer1.1.bn1.bias : torch.Size([64])
layer1.1.conv2.weight : torch.Size([64, 64, 3, 3])
layer1.1.bn2.weight : torch.Size([64])
layer1.1.bn2.bias : torch.Size([64])
layer2.0.conv1.weight : torch.Size([128, 64, 3, 3])
layer2.0.bn1.weight : torch.Size([128])
layer2.0.bn1.bias : torch.Size([128])
layer2.0.conv2.weight : torch.Size([128, 128, 3, 3])
layer2.0.bn2.weight : torch.Size([128])
layer2.0.bn2.bias : torch.Size([128])
layer2.0.downsample.0.weight : torch.Size([128, 64, 1, 1])
layer2.0.downsample.1.weight : torch.Size([128])
layer2.0.downsample.1.bias : torch.Size([128])
layer2.1.conv1.weight : torch.Size([128, 128, 3, 3])
layer2.1.bn1.weight : torch.Size([128])
layer2.1.bn1.bias : torch.Size([128])
layer2.1.conv2.weight : torch.Size([128, 128, 3, 3])
layer2.1.bn2.weight : torch.Size([128])
layer2.1.bn2.bias : torch.Size([128])
layer3.0.conv1.weight : torch.Size([256, 128, 3, 3])
layer3.0.bn1.weight : torch.Size([256])
layer3.0.bn1.bias : torch.Size([256])
layer3.0.conv2.weight : torch.Size([256, 256, 3, 3])
layer3.0.bn2.weight : torch.Size([256])
layer3.0.bn2.bias : torch.Size([256])
layer3.0.downsample.0.weight : torch.Size([256, 128, 1, 1])
layer3.0.downsample.1.weight : torch.Size([256])
layer3.0.downsample.1.bias : torch.Size([256])
layer3.1.conv1.weight : torch.Size([256, 256, 3, 3])
layer3.1.bn1.weight : torch.Size([256])
layer3.1.bn1.bias : torch.Size([256])
layer3.1.conv2.weight : torch.Size([256, 256, 3, 3])
layer3.1.bn2.weight : torch.Size([256])
layer3.1.bn2.bias : torch.Size([256])
layer4.0.conv1.weight : torch.Size([512, 256, 3, 3])
layer4.0.bn1.weight : torch.Size([512])
layer4.0.bn1.bias : torch.Size([512])
layer4.0.conv2.weight : torch.Size([512, 512, 3, 3])
layer4.0.bn2.weight : torch.Size([512])
layer4.0.bn2.bias : torch.Size([512])
layer4.0.downsample.0.weight : torch.Size([512, 256, 1, 1])
layer4.0.downsample.1.weight : torch.Size([512])
layer4.0.downsample.1.bias : torch.Size([512])
layer4.1.conv1.weight : torch.Size([512, 512, 3, 3])
layer4.1.bn1.weight : torch.Size([512])
layer4.1.bn1.bias : torch.Size([512])
layer4.1.conv2.weight : torch.Size([512, 512, 3, 3])
layer4.1.bn2.weight : torch.Size([512])
layer4.1.bn2.bias : torch.Size([512])
fc.weight : torch.Size([1000, 512])
fc.bias : torch.Size([1000])

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9
• 10
• 11
• 12
• 13
• 14
• 15
• 16
• 17
• 18
• 19
• 20
• 21
• 22
• 23
• 24
• 25
• 26
• 27
• 28
• 29
• 30
• 31
• 32
• 33
• 34
• 35
• 36
• 37
• 38
• 39
• 40
• 41
• 42
• 43
• 44
• 45
• 46
• 47
• 48
• 49
• 50
• 51
• 52
• 53
• 54
• 55
• 56
• 57
• 58
• 59
• 60
• 61
• 62

parm['layer1.0.conv1.weight'][0,0,:,:]

• 1

输出为:

array([[ 0.05759342, -0.09511436, -0.02027232],
       [-0.07455588, -0.799308  , -0.21283598],
       [ 0.06557069, -0.09653367, -0.01211061]], dtype=float32)

• 1
• 2
• 3

利用如下函数将某一层的所有参数保存到表格中,数据维持卷积核特征大小,如3*3的卷积保存后还是3x3的.

def parm_to_excel(excel_name,key_name,parm):
    with pd.ExcelWriter(excel_name) as writer:
        [output_num,input_num,filter_size,_]=parm[key_name].size()
        for i in range(output_num):
            for j in range(input_num):
                data=pd.DataFrame(parm[key_name][i,j,:,:].detach().numpy())
                #print(data)
                data.to_excel(writer,index=False,header=True,startrow=i*(filter_size+1),startcol=j*filter_size)

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8

由于权重矩阵中有很多的值非常小,取出固定大小的值,并将全部权重写入excel

counter=1
with pd.ExcelWriter('test1.xlsx') as writer:
    for key in parm_resnet50.keys():
        data=parm_resnet50[key].reshape(-1,1)
        data=data[data>0.001]
        
        data=pd.DataFrame(data,columns=[key])
        data.to_excel(writer,index=False,startcol=counter)
        counter+=1

• 1
• 2
• 3
• 4
• 5
• 6
• 7
• 8
• 9

2.获取中间某一层的特性

重写一个函数,将需要输出的层输出即可.

def resnet_cifar(net,input_data):
    x = net.conv1(input_data)
    x = net.bn1(x)
    x = F.relu(x)
    x = net.layer1(x)
    x = net.layer2(x)
    x = net.layer3(x)
    x = net.layer4[0].conv1(x)  #这样就提取了layer4第一块的第一个卷积层的输出
    x=x.view(x.shape[0],-1)
    return x

model = models.resnet18()
x = resnet_cifar(model,input_data)