def get_model(hidden_size: int, num_layers: int, enc_dropout: float,
enc_emb_dropout: float, dec_dropout: float,
dec_emb_dropout: float, seq_enc_dropout,
seq_dec_dropout: float) -> PtrNetModel:
form_emb = model_ft.load_embedding_weight_matrix(home_path,
ft_form_vec_file_path,
tb_vocab.forms, device)
lemma_emb = model_ft.load_embedding_weight_matrix(home_path,
ft_lemma_vec_file_path,
tb_vocab.lemmas, device)
tags_num = len(tb_vocab.tags)
feats_num = len(tb_vocab.feats)
tag_emb = nn.Embedding(num_embeddings=tags_num,
embedding_dim=50,
padding_idx=0)
feats_emb = nn.Embedding(num_embeddings=feats_num,
embedding_dim=50,
padding_idx=0)
analysis_emb = AnalysisEmbedding(form_emb, lemma_emb, tag_emb, feats_emb)
encoder = AnalysisEncoder(enc_dropout, analysis_emb.embedding_dim,
hidden_size, num_layers, enc_emb_dropout)
decoder = AnalysisDecoder(dec_dropout, analysis_emb.embedding_dim,
hidden_size, num_layers, dec_emb_dropout)
attention = Attention()
model = PtrNetModel(tb_vocab, analysis_emb, encoder, seq_enc_dropout,
decoder, seq_dec_dropout, attention)
if torch.cuda.is_available():
model.cuda(device)
return model
def get_model(hidden_size: int, num_layers: int, emb_dropout: float,
hidden_dropout: float, rnn_dropout: float,
class_dropout: float) -> TagRNN4:
token_emb = model_ft.load_embedding_weight_matrix(home_path,
ft_token_vec_file_path,
tb_vocab.tokens, device)
model = TagRNN4(token_emb, emb_dropout, hidden_size, num_layers,
hidden_dropout, rnn_dropout, class_dropout, tb_vocab)
if torch.cuda.is_available():
model.cuda(device)
return model