RNN初探

最新推荐文章于 2023-05-03 06:49:46 发布

self.键盘上的钢琴师.LX

最新推荐文章于 2023-05-03 06:49:46 发布

阅读量163

点赞数

文章标签： rnn lstm 深度学习

本文链接：https://blog.youkuaiyun.com/lixiangohoh/article/details/124839430

版权

该博客演示了如何使用PyTorch构建并训练一个简单的RNN模型。首先，导入必要的库和数据集，然后定义RNN模型，接着通过RNN层处理输入数据。在模型中添加全连接层，并实现前向传播方法。最后，对模型进行预测和训练。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

import torch
from torch import nn
from torch.nn import functional as F
from d2l import torch as d2l
# 导入数据
batch_size, num_step = 32, 35
train_iter, vocab = d2l.load_data_time_machine(batch_size, num_step)
# 定义模型
num_hiddens = 256
rnn_layer = nn.RNN(len(vocab), num_hiddens)
state = torch.zeros((1, batch_size, num_hiddens))
# 隐藏层数，批量大小，隐藏单元数
# print(state.shape)
x = torch.rand(size=(num_step, batch_size, len(vocab)))
y, state_new = rnn_layer(x, state)
print(y.shape, state_new.shape)
# 定义循环神经网络的类，并增加输出层
class RNNModel(nn.Module):
    '''循环神经网络'''
    def __init__(self, rnn_layer, vocab_size):
        super(RNNModel, self).__init__()
        self.rnn = rnn_layer
        self.vocab_size = vocab_size
        self.num_hiddens = self.rnn.hidden_size
        # 如果rnn为双向则，num_directions为2，否则为1
        if not self.rnn.bidirectional:
            self.num_directions = 1
            self.linear = nn.Linear(self.num_hiddens, self.vocab_size)
        else:
            self.num_directions = 2
            self.linear = nn.Linear(self.num_hiddens * 2, self.vocab_size)

    def forward(self, inputs, state):
        X = F.one_hot(inputs.T.long(), self.vocab_size)
        X = X.to(torch.float32)
        Y, state = self.rnn(X, state)
        # 全连接层首先将Y的形状改为（时间步数*批量大小， 隐藏单元个数）
        # 它的输出形状是（时间步数*批量大小， 词表大小）
        output = self.linear(Y.reshape((-1, Y.shape[-1])))
        return output, state

    def begin_state(self, device, batch_size):
        if not isinstance(self.rnn, nn.LSTM):
            # nn.GRU以张量作为隐藏状态
            return torch.zeros((self.num_directions * self.rnn.num_layers, batch_size, self.num_hiddens), device=device)
        else:
            # nn.LSTM以元组作为隐藏状态
            return (torch.zeros((
                self.num_directions * self.rnn.num_layers,
                batch_size, self.num_hiddens), device=device),
                torch.zeros((self.num_directions * self.rnn.num_layers, batch_size, self.num_hiddens), device=device))


if __name__ == '__main__':
    device = d2l.try_gpu()
    net = RNNModel(rnn_layer, vocab_size=len(vocab))
    net = net.to(device)
    no_train = d2l.predict_ch8('my name is lixiang', 10, net, vocab, device)
    print(no_train)

    # 对网络进行训练
    num_epoch, lr = 1000, 0.99
    result = d2l.train_ch8(net, train_iter, vocab, lr, num_epoch, device)
    print(result)