python - 如何在循环中为 pytorch 神经网络中的层创建变量名称

Question

我正在 PyTorch 中实现一个简单的前馈神经 newtork。但是我想知道是否有更好的方法来向网络添加灵活的层数？也许通过在循环中命名它们，但我听说那是不可能的？

目前我正在这样做

import torch
import torch.nn as nn
import torch.nn.functional as F

class Net(nn.Module):

    def __init__(self, input_dim, output_dim, hidden_dim):
        super(Net, self).__init__()
        self.input_dim = input_dim
        self.output_dim = output_dim
        self.hidden_dim = hidden_dim
        self.layer_dim = len(hidden_dim)
        self.fc1 = nn.Linear(self.input_dim, self.hidden_dim[0])
        i = 1
        if self.layer_dim > i:
            self.fc2 = nn.Linear(self.hidden_dim[i-1], self.hidden_dim[i])
            i += 1
        if self.layer_dim > i:
            self.fc3 = nn.Linear(self.hidden_dim[i-1], self.hidden_dim[i])
            i += 1
        if self.layer_dim > i:
            self.fc4 = nn.Linear(self.hidden_dim[i-1], self.hidden_dim[i])
            i += 1
        if self.layer_dim > i:
            self.fc5 = nn.Linear(self.hidden_dim[i-1], self.hidden_dim[i])
            i += 1
        if self.layer_dim > i:
            self.fc6 = nn.Linear(self.hidden_dim[i-1], self.hidden_dim[i])
            i += 1
        if self.layer_dim > i:
            self.fc7 = nn.Linear(self.hidden_dim[i-1], self.hidden_dim[i])
            i += 1
        if self.layer_dim > i:
            self.fc8 = nn.Linear(self.hidden_dim[i-1], self.hidden_dim[i])
            i += 1
        self.fcn = nn.Linear(self.hidden_dim[-1], self.output_dim)

    def forward(self, x):
        # Max pooling over a (2, 2) window
        x = F.relu(self.fc1(x))
        i = 1
        if self.layer_dim > i:
            x = F.relu(self.fc2(x))
            i += 1
        if self.layer_dim > i:
            x = F.relu(self.fc3(x))
            i += 1
        if self.layer_dim > i:
            x = F.relu(self.fc4(x))
            i += 1
        if self.layer_dim > i:
            x = F.relu(self.fc5(x))
            i += 1
        if self.layer_dim > i:
            x = F.relu(self.fc6(x))
            i += 1
        if self.layer_dim > i:
            x = F.relu(self.fc7(x))
            i += 1
        if self.layer_dim > i:
            x = F.relu(self.fc8(x))
            i += 1
        x = F.softmax(self.fcn(x))
        return x

Answer 1

您可以将图层放在 ModuleList 容器中:

import torch
import torch.nn as nn
import torch.nn.functional as F

class Net(nn.Module):

    def __init__(self, input_dim, output_dim, hidden_dim):
        super(Net, self).__init__()
        self.input_dim = input_dim
        self.output_dim = output_dim
        self.hidden_dim = hidden_dim
        current_dim = input_dim
        self.layers = nn.ModuleList()
        for hdim in hidden_dim:
            self.layers.append(nn.Linear(current_dim, hdim))
            current_dim = hdim
        self.layers.append(nn.Linear(current_dim, output_dim))

    def forward(self, x):
        for layer in self.layers[:-1]:
            x = F.relu(layer(x))
        out = F.softmax(self.layers[-1](x))
        return out    

对层使用 pytorch Containers 非常重要，而不仅仅是简单的 Python 列表。请参阅 this answer 以了解原因。