RNN的幾種結構

• Elman RNN

•  Jordan RNN

 

• Bidirectional RNN

• LSTM - Cell

 

LSTM：

理解Pytorch中的LSTM （refer:https://www.cnblogs.com/marsggbo/p/12123755.html）

 

lstm=nn.LSTM(input_size=300,hidden_size=128,num_layers=2)

input=torch.rand([10,64,300]) #seq_len, batch, input_size
h0=torch.rand([2,64,128]) # num_layes * direction_numbers,batch,hidden_size
c0=torch.rand([2,64,128]) # num_layes * direction_numbers,batch,hidden_size
c

output,(hn,cn)=lstm(input,(h0,c0))
print(output.shape) # seq_len,batch,hidden_size
print(hn.shape) # num_layes * direction_numbers,batch,hidden_size
print(cn.shape) # num_layes * direction_numbers,batch,hidden_size

'''
torch.Size([10, 64, 128])
torch.Size([2, 64, 128])
torch.Size([2, 64, 128])
'''