本文整理匯總了Python中onmt.modules.Elementwise方法的典型用法代碼示例。如果您正苦於以下問題：Python modules.Elementwise方法的具體用法？Python modules.Elementwise怎麽用？Python modules.Elementwise使用的例子？那麽, 這裏精選的方法代碼示例或許可以為您提供幫助。您也可以進一步了解該方法所在類onmt.modules的用法示例。
在下文中一共展示了modules.Elementwise方法的1個代碼示例，這些例子默認根據受歡迎程度排序。您可以為喜歡或者感覺有用的代碼點讚，您的評價將有助於係統推薦出更棒的Python代碼示例。
示例1: __init__
# 需要導入模塊: from onmt import modules [as 別名]
# 或者: from onmt.modules import Elementwise [as 別名]
def __init__(self, word_vec_size, position_encoding, feat_merge,
feat_vec_exponent, feat_vec_size, dropout,
word_padding_idx, feat_padding_idx,
word_vocab_size, feat_vocab_sizes=[]):
self.word_padding_idx = word_padding_idx
# Dimensions and padding for constructing the word embedding matrix
vocab_sizes = [word_vocab_size]
emb_dims = [word_vec_size]
pad_indices = [word_padding_idx]
# Dimensions and padding for feature embedding matrices
# (these have no effect if feat_vocab_sizes is empty)
if feat_merge == 'sum':
feat_dims = [word_vec_size] * len(feat_vocab_sizes)
elif feat_vec_size > 0:
feat_dims = [feat_vec_size] * len(feat_vocab_sizes)
else:
feat_dims = [int(vocab ** feat_vec_exponent)
for vocab in feat_vocab_sizes]
vocab_sizes.extend(feat_vocab_sizes)
emb_dims.extend(feat_dims)
pad_indices.extend(feat_padding_idx)
# The embedding matrix look-up tables. The first look-up table
# is for words. Subsequent ones are for features, if any exist.
emb_params = zip(vocab_sizes, emb_dims, pad_indices)
embeddings = [nn.Embedding(vocab, dim, padding_idx=pad)
for vocab, dim, pad in emb_params]
emb_luts = Elementwise(feat_merge, embeddings)
# The final output size of word + feature vectors. This can vary
# from the word vector size if and only if features are defined.
# This is the attribute you should access if you need to know
# how big your embeddings are going to be.
self.embedding_size = (sum(emb_dims) if feat_merge == 'concat'
else word_vec_size)
# The sequence of operations that converts the input sequence
# into a sequence of embeddings. At minimum this consists of
# looking up the embeddings for each word and feature in the
# input. Model parameters may require the sequence to contain
# additional operations as well.
super(Embeddings, self).__init__()
self.make_embedding = nn.Sequential()
self.make_embedding.add_module('emb_luts', emb_luts)
if feat_merge == 'mlp':
in_dim = sum(emb_dims)
out_dim = word_vec_size
mlp = nn.Sequential(BottleLinear(in_dim, out_dim), nn.ReLU())
self.make_embedding.add_module('mlp', mlp)
if position_encoding:
pe = PositionalEncoding(dropout, self.embedding_size)
self.make_embedding.add_module('pe', pe)