用LSTM来做下那个预测问题

LSTM要初始化的参数：input_size,hidden_size,num_layers,batch_first

import torch

class Model(torch.nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.emb = torch.nn.Embedding(input_size, embedding_size)
        self.lstm = torch.nn.LSTM(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)
        # 因为LSTM输入限制，要多弄一个c的输入，和hidden一起放进去
        c = torch.zeros(num_layers, x.size(0), hidden_size)
        x = self.emb(x) #这样就可以变成稠密的向量
        x, _ = self.lstm(x, (hidden, c))
        x = self.fc(x)
        return x.view(-1,num_class)

#一堆参数
num_class = 4
input_size = 4
hidden_size = 8
embedding_size = 10
num_layers = 2
batch_size = 1
seq_len = 5

#准备数据
idx2char = ['e', 'h', 'l', 'o']
x_data = [[1, 0, 2, 2, 3]] # (batch, seq_len)
y_data = [3, 1, 2, 3, 2] # (batch * seq_len)

inputs = torch.LongTensor(x_data).view(batch_size, seq_len)
labels = torch.LongTensor(y_data)

#实例化模型并构造损失函数和优化器
net = Model()

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

#训练
for epoch in range(20):
    optimizer.zero_grad()
    outputs = net(inputs)
    loss = criterion(outputs, labels)
    loss.backward()
    optimizer.step()

    _, idx = outputs.max(dim=1)
    idx = idx.data.numpy()
    print('Predicted string:', ''.join([idx2char[x] for x in idx]), end='')
    print(', Epoch[%d/20] loss=%.4f' % (epoch+1, loss.item()))

注意，LSTM的输入是input,(h_0,c_0)，所以要再多弄一个c_0

用GRU做

GRU要初始化的参数：input_size,hidden_size

import torch

class Model(torch.nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.emb = torch.nn.Embedding(input_size, embedding_size)
        self.gru = torch.nn.GRU(embedding_size, hidden_size)
        self.fc = torch.nn.Linear(hidden_size, num_class)

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

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

idx2char = ['e', 'h', 'l', 'o']
x_data = [1, 0, 2, 2, 3] 
y_data = [3, 1, 2, 3, 2] 

inputs = torch.LongTensor(x_data).view(batch_size, seq_len)
labels = torch.LongTensor(y_data)

net = Model()

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

for epoch in range(20):
    optimizer.zero_grad()
    outputs = net(inputs)
    loss = criterion(outputs, labels)
    loss.backward()
    optimizer.step()

    _, idx = outputs.max(dim=1)
    idx = idx.data.numpy()
    print('Predicted string:', ''.join([idx2char[x] for x in idx]), end='')
    print(', Epoch[%d/20] loss=%.4f' % (epoch+1, loss.item()))

这里不知道为啥要改成x.size(1)，而且最后的结果都不是想要的结果。。。