【Pytorch错误】RNN-GRU代码编写，出现didn‘t match because some of the arguments have invalid types: (float, int)

63 阅读 0 评论 42 点赞

我是靠谱客的博主靓丽白昼，最近开发中收集的这篇文章主要介绍【Pytorch错误】RNN-GRU代码编写，出现didn‘t match because some of the arguments have invalid types: (float, int)，觉得挺不错的，现在分享给大家，希望可以做个参考。

概述

在使用RNN-GRU模块编写时出现：

didn't match because some of the arguments have invalid types: (float, int)

这是因为，RNN的隐藏层个数除以2到GRU中出现了数据类型不匹配，hidden_size=configs.hidden_size /2为浮点型，需要改为整型数据，即错误代码：

import torch
import torch.nn as nn

class RNN_GRU(nn.Module):
    def __init__(self, configs):
        super(RNN_GRU, self).__init__()
        self.hidden_size = configs.hidden_size
        self.num_layers = configs.num_layers
        self.seq_len = configs.seq_len
        self.pred_len = configs.pred_len
        self.rnn = nn.RNN(input_size=1, hidden_size=configs.hidden_size, num_layers=configs.num_layers,
                          batch_first=True)  # batch_first – 如果为True，那么输入和输出Tensor的形状为(batch, seq, feature)
        self.gru = nn.GRU(input_size=configs.hidden_size, hidden_size=configs.hidden_size /2,
                          num_layers=configs.num_layers,
                          batch_first=True)  # batch_first – 如果为True，那么输入和输出Tensor的形状为(batch, seq, feature)
        self.fc = nn.Linear(configs.hidden_size / 2, configs.pred_len)

    def forward(self, x):
        # Set initial hidden and cell states
        h0 = torch.zeros(self.num_layers, x.size(0), self.hidden_size).cuda()
        out, _ = self.rnn(x, h0)  # out: tensor of shape (batch_size, seq_length, hidden_size)
        h1 = torch.zeros(self.num_layers, out.size(0), self.hidden_size / 2).cuda()
        out, _ = self.gru(out, h1)  # out: tensor of shape (batch_size, seq_length, hidden_size)
        out = self.fc(out[:, -1, :])  # 此处的-1说明我们只取RNN最后输出的那个hn
        return out

正确的做法是将’/‘全部改为’//'，正确代码如下：

import torch
import torch.nn as nn

class RNN_GRU(nn.Module):
    def __init__(self, configs):
        super(RNN_GRU, self).__init__()
        self.hidden_size = configs.hidden_size
        self.num_layers = configs.num_layers
        self.seq_len = configs.seq_len
        self.pred_len = configs.pred_len
        self.rnn = nn.RNN(input_size=1, hidden_size=configs.hidden_size, num_layers=configs.num_layers,
                          batch_first=True)  # batch_first – 如果为True，那么输入和输出Tensor的形状为(batch, seq, feature)
        self.gru = nn.GRU(input_size=configs.hidden_size, hidden_size=configs.hidden_size //2,
                          num_layers=configs.num_layers,
                          batch_first=True)  # batch_first – 如果为True，那么输入和输出Tensor的形状为(batch, seq, feature)
        self.fc = nn.Linear(configs.hidden_size // 2, configs.pred_len)

    def forward(self, x):
        # Set initial hidden and cell states
        h0 = torch.zeros(self.num_layers, x.size(0), self.hidden_size).cuda()
        out, _ = self.rnn(x, h0)  # out: tensor of shape (batch_size, seq_length, hidden_size)
        h1 = torch.zeros(self.num_layers, out.size(0), self.hidden_size // 2).cuda()
        out, _ = self.gru(out, h1)  # out: tensor of shape (batch_size, seq_length, hidden_size)
        out = self.fc(out[:, -1, :])  # 此处的-1说明我们只取RNN最后输出的那个hn
        return out