def build_embeddings(opt, word_dict, for_encoder='src'):
"""
Build an Embeddings instance.
Args:
opt: the option in current environment.
word_dict(Vocab): words dictionary.
feature_dicts([Vocab], optional): a list of feature dictionary.
for_encoder(bool): build Embeddings for encoder or decoder?
"""
if for_encoder == 'src':
embedding_dim = opt.src_word_vec_size # 512
elif for_encoder == 'tgt':
embedding_dim = opt.tgt_word_vec_size
elif for_encoder == 'structure':
embedding_dim = 64
word_padding_idx = word_dict.stoi[Constants.PAD_WORD]
num_word_embeddings = len(word_dict)
if for_encoder == 'src' or for_encoder == 'tgt':
return Embeddings(word_vec_size=embedding_dim,
position_encoding=opt.position_encoding,
dropout=opt.dropout,
word_padding_idx=word_padding_idx,
word_vocab_size=num_word_embeddings,
sparse=opt.optim == "sparseadam")
elif for_encoder == 'structure':
return Embeddings(word_vec_size=embedding_dim,
position_encoding=False,
dropout=opt.dropout,
word_padding_idx=word_padding_idx,
word_vocab_size=num_word_embeddings,
sparse=opt.optim == "sparseadam")
def build_embeddings(opt, word_dict, for_encoder=True):
"""
Build an Embeddings instance.
Args:
opt: the option in current environment.
word_dict(Vocab): words dictionary.
feature_dicts([Vocab], optional): a list of feature dictionary.
for_encoder(bool): build Embeddings for encoder or decoder?
"""
if for_encoder:
# word_embedding大小默认是512
embedding_dim = opt.src_word_vec_size
else:
embedding_dim = opt.tgt_word_vec_size
# 获取填充数值 1
word_padding_idx = word_dict.stoi[Constants.PAD_WORD]
# 获取vocab大小
num_word_embeddings = len(word_dict)
return Embeddings(word_vec_size=embedding_dim,
dropout=opt.dropout,
word_padding_idx=word_padding_idx,
word_vocab_size=num_word_embeddings,
sparse=opt.optim == "sparseadam")