QLSTM POS Tagger

qlstm_pos_tagger.py

class LSTMTagger(nn.Module):
    def __init__(self, embedding_dim, hidden_dim, vocab_size, tagset_size, n_qubits=0):
        super(LSTMTagger, self).__init__()
        self.hidden_dim = hidden_dim

        self.word_embeddings = nn.Embedding(vocab_size, embedding_dim)

        # The LSTM takes word embeddings as inputs, and outputs hidden states
        # with dimensionality hidden_dim.
        if n_qubits > 0:
            print("Tagger will use Quantum LSTM")
            self.lstm = QLSTM(embedding_dim, hidden_dim, n_qubits=n_qubits)
        else:
            print("Tagger will use Classical LSTM")
            self.lstm = nn.LSTM(embedding_dim, hidden_dim)

        # The linear layer that maps from hidden state space to tag space
        self.hidden2tag = nn.Linear(hidden_dim, tagset_size)

    def forward(self, sentence):
        embeds = self.word_embeddings(sentence)
        lstm_out, _ = self.lstm(embeds.view(len(sentence), 1, -1))
        tag_logits = self.hidden2tag(lstm_out.view(len(sentence), -1))
        tag_scores = F.log_softmax(tag_logits, dim=1)
        return tag_scores

loss_function = nn.NLLLoss(). # the output is a log_softmax!
optimizer = optim.SGD(model.parameters(), lr=0.1)