Pytorch实现RNN

Pytorch实现RNN：

一、torch.nn.RNN()

1. 导入所需包

import torch
from torch.autograd import Variable # 获取变量
from torch.utils import data # 获取迭代数
import torchvision 
from torchvision.datasets import mnist # 获取数据集
1
2
3
4
5

2. 获取数据

2.1定义数据集预处理的方法

# 预处理
data_tf = torchvision.transforms.Compose(
    [torchvision.transforms.ToTensor(),
     torchvision.transforms.Normalize([0.5],[0.5])]
)
1
2
3
4
5

2.2 获取数据，同时对其进行预处理

# 获取数据
data_path = r'C:\Users\liev\Desktop\myproject\yin_test\MNIST_DATA_PyTorch'
train_data = mnist.MNIST(data_path,train=True,transform=data_tf,download=False)
test_data = mnist.MNIST(data_path,train=False,transform=data_tf,download=False)
1
2
3
4

3.网络结构

# 定义网络结构
class RNNnet(torch.nn.Module):
    def __init__(self):
        super(RNNnet, self).__init__()
        self.rnn1 = torch.nn.RNN(784,100,3,nonlinearity='relu')
        self.rnn2 = torch.nn.RNN(100,10,1,nonlinearity='relu')
    def forward(self, x):
        x = self.rnn1(x)
        x = torch.Tensor(x[1])
        x = self.rnn2(
1
2
3
4
5
6
7
8
9