Advanced Deep Learning with Python

Implementing the decoder

Let's implement the DecoderRNN class—a basic decoder without attention. Again, we'll start with the constructor:

class DecoderRNN(torch.nn.Module):    def __init__(self, hidden_size, output_size):        super(DecoderRNN, self).__init__()        self.hidden_size = hidden_size        self.output_size = output_size        # Embedding for the current input word        self.embedding = torch.nn.Embedding(output_size, hidden_size)        # decoder cell        self.gru = torch.nn.GRU(hidden_size, hidden_size)        # Current output word        self.out = torch.nn.Linear(hidden_size, output_size)        self.log_softmax = torch.nn.LogSoftmax(dim=1)

It's similar to the encoder—we have the initial self.embedding word embedding and the self.gru GRU cell. We also have the fully connected self.out ...