pytorch utiliza la visualización del modelo de aprendizaje profundo de capa oculta

Descripción del problema

Al escribir un modelo de red neuronal, a menudo es imposible juzgar intuitivamente si el modelo que diseñamos es lo que imaginamos, por lo que es necesario visualizar el modelo.

solución:

1. Instale la capa oculta:

 pip install hiddenlayer

2. Ejemplos de aplicación

Por ejemplo, hay una red neuronal de dos capas

class CNN2(nn.Module):  #3 layers
    def __init__(self,num_classes=10):
        super(CNN2, self).__init__()

        self.layer1 = nn.Sequential(
            nn.Conv2d(1, 16, kernel_size=3),  # 16, 26 ,26
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True)
        )
        self.layer2 = nn.Sequential(
            nn.Conv2d(16, 16, kernel_size=1),  # 16, 26 ,26
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True)
        )
        

        
        self.fc1 = nn.Linear(1*10816, 2048)
        self.fc2 = nn.Linear(2048, num_classes)

    def forward(self, x):
        x = self.layer1(x)
        x = self.layer2(x)
        x = x.view(x.size(0), -1)
        x = self.fc1(x)
        x = self.fc2(x)

        return F.log_softmax(x, dim=1)

Funciones personalizadas para imprimir parámetros del modelo y visualizar imágenes

def modelhiddenlayer(model,input,input2):
     # import hiddenlayer as h
     # params: model = MSDNet(args).cuda() input = (3, 32, 32) input2 = [1 ,3, 32,32]
     print(model)        
     summary(model, input)
     vis_graph = h.build_graph(model, torch.zeros(input2).cuda())   # 获取绘制图像的对象
     vis_graph.theme = h.graph.THEMES["blue"].copy()     # 指定主题颜色
     vis_graph.save("./demo1.png")   # 保存图像的路径

definir la función principal

def main():
    model = CNN2(10).cuda()
    input = (1, 28, 28) 
    input2 = [1,1,28,28]
    modelhiddenlayer(model,input,input2) 
  
if __name__ == '__main__':
    main()

El resultado se muestra en la figura.

¡éxito!
código completo

import torch
from torch import nn
import torch.nn.functional as F
from torchsummary import summary
import hiddenlayer as h
from torchviz import make_dot

class CNN2(nn.Module):  #3 layers
    def __init__(self,num_classes=10):
        super(CNN2, self).__init__()

        self.layer1 = nn.Sequential(
            nn.Conv2d(1, 16, kernel_size=3),  # 16, 26 ,26
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True)
        )
        self.layer2 = nn.Sequential(
            nn.Conv2d(16, 16, kernel_size=1),  # 16, 26 ,26
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True)
        )
        

        
        self.fc1 = nn.Linear(1*10816, 2048)
        self.fc2 = nn.Linear(2048, num_classes)

    def forward(self, x):
        x = self.layer1(x)
        x = self.layer2(x)
        x = x.view(x.size(0), -1)
        x = self.fc1(x)
        x = self.fc2(x)

        return F.log_softmax(x, dim=1)

def modelhiddenlayer(model,input,input2):
     # import hiddenlayer as h
     # params: model = MSDNet(args).cuda() input = (3, 32, 32) input2 = [1 ,3, 32,32]
     print(model)        
     summary(model, input)
     vis_graph = h.build_graph(model, torch.zeros(input2).cuda())   # 获取绘制图像的对象
     vis_graph.theme = h.graph.THEMES["blue"].copy()     # 指定主题颜色
     vis_graph.save("./demo1.png")   # 保存图像的路径 
    
def main():   
    model = CNN2(10).cuda()
    input = (1, 28,28) 
    input2 = [1 ,1,28,28]
    modelhiddenlayer(model,input,input2) 
    
if __name__ == '__main__':
    main()