EmbeddingRNN

RNN–Embedding and Linear Layer

• 目标

• 网络总体框架

import torch

idx2char = ['e', 'h', 'l', 'o']
x_data = [[1, 0, 2, 2, 3]]  # The input sequence is 'hello'  (batch, seq_len)，不同于charater_testRNN 和 BasicRNN 中的方式
y_data = [3, 1, 2, 3, 2]  # The output sequence is 'ohlol'   (batch * seq_len)

# Embedding层要求 input 和 target 是 LongTensor
inputs = torch.LongTensor(x_data)
labels = torch.LongTensor(y_data)

num_class = 4
input_size = 4
hidden_size = 8
embedding_size = 10
num_layers = 2
batch_size = 1
seq_len = 5

class Model(torch.nn.Module):

    def __init__(self):
        super(Model, self).__init__()
        self.emb = torch.nn.Embedding(input_size, embedding_size)
        self.rnn = torch.nn.RNN(input_size=embedding_size, hidden_size=hidden_size, 
                                num_layers=num_layers, batch_first=True)
        self.fc = torch.nn.Linear(hidden_size, num_class) 

    def forward(self, x):
        hidden = torch.zeros(num_layers, x.size(0), hidden_size)
        x = self.emb(x)
        x, _ = self.rnn(x, hidden)
        x = self.fc(x)
        return x.view(-1, num_class)                           

net = Model()
criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(net.parameters(), lr=0.05)

for epoch in range(15):
    optimizer.zero_grad()
    outputs = net(inputs)
    loss = criterion(outputs, labels)
    loss.backward()
    optimizer.step()

    _, idx = outputs.max(dim=1)
    idx = idx.data.numpy()  # 数据转换成 numpy 格式
    print('Predicted: ', ''.join([idx2char[x] for x in idx]), end='')
    print(', Epocn [%d / 15] loss = %.3f' % (epoch + 1, loss.item()))

运行结果

    Predicted:  ellel, Epocn [1 / 15] loss = 1.532
    Predicted:  lllll, Epocn [2 / 15] loss = 1.202
    Predicted:  ohlll, Epocn [3 / 15] loss = 0.972
    Predicted:  ohlll, Epocn [4 / 15] loss = 0.763
    Predicted:  ohlol, Epocn [5 / 15] loss = 0.593
    Predicted:  ohlol, Epocn [6 / 15] loss = 0.439
    Predicted:  ohlol, Epocn [7 / 15] loss = 0.312
    Predicted:  ohlol, Epocn [8 / 15] loss = 0.217
    Predicted:  ohlol, Epocn [9 / 15] loss = 0.151
    Predicted:  ohlol, Epocn [10 / 15] loss = 0.105
    Predicted:  ohlol, Epocn [11 / 15] loss = 0.074
    Predicted:  ohlol, Epocn [12 / 15] loss = 0.053
    Predicted:  ohlol, Epocn [13 / 15] loss = 0.039
    Predicted:  ohlol, Epocn [14 / 15] loss = 0.030
    Predicted:  ohlol, Epocn [15 / 15] loss = 0.024