def get_embeddings(hparams):
if hparams.glove_path and hparams.vocab_path:
tf.logging.info("Loading Glove embeddings...")
vocab_array, vocab_dict = helpers.load_vocab(hparams.vocab_path)
glove_vectors, glove_dict = helpers.load_glove_vectors(
hparams.glove_path, vocab=set(vocab_array))
initializer = helpers.build_initial_embedding_matrix(
vocab_dict, glove_dict, glove_vectors, hparams.embedding_dim)
else:
tf.logging.info(
"No glove/vocab path specificed, starting with random embeddings.")
initializer = tf.random_uniform_initializer(-0.25, 0.25)
if hparams.glove_path and hparams.vocab_path:
return tf.get_variable("word_embeddings", initializer=initializer)
elif hparams.vocab_path:
vocab_array, vocab_dict = helpers.load_vocab(hparams.vocab_path)
return tf.get_variable("word_embeddings",
shape=[len(vocab_dict), hparams.embedding_dim],
initializer=initializer)
else:
return tf.get_variable(
"word_embeddings",
shape=[hparams.vocab_size, hparams.embedding_dim],
initializer=initializer)
def get_embeddings(hparams):
vocab_array, vocab_dict = helpers.load_vocab(hparams.vocab_path)
print("vacab_array / dict loaded.")
glove_vectors, glove_dict = helpers.load_glove_vectors(
hparams.glove_path, vocab=set(vocab_array))
print("glove_vectors / dict loaded.")
W = helpers.build_initial_embedding_matrix(vocab_dict, glove_dict,
glove_vectors,
hparams.embedding_dim)
print("Embedding matrix built.")
return W
def glove_init(embedding_size):
"""GloVe initialization"""
glove_path = "data/glove.6B.100d.txt"
vocab_path = "data/vocabulary.txt"
tf.logging.info("Loading GloVe embeddings ...")
vocab_array, vocab_dict = helpers.load_vocab(vocab_path)
glove_vectors, glove_dict = helpers.load_glove_vectors(
glove_path, vocab=set(vocab_array))
initializer = helpers.build_initial_embedding_matrix(
vocab_dict, glove_dict, glove_vectors, embedding_size)
return initializer
def get_embeddings(hparams):
# 加载词汇表，训练数据中包含的
vocab_array, vocab_dict = helpers.load_vocab(hparams.vocab_path)
if hparams.vector_type == 'word2vec':
word2vec_vectors, word2vec_dict = helpers.load_word2vec_vectors(
hparams.word2vec_path, vocab=set(vocab_array))
initializer = helpers.build_initial_embedding_matrix(
vocab_dict, word2vec_dict, word2vec_vectors, hparams.embedding_dim)
elif hparams.vector_type == 'glove':
# glove_vectors所有出现此的向量； glove_dict记录出现此的位置
glove_vectors, glove_dict = helpers.load_glove_vectors(
hparams.glove_path, vocab=set(vocab_array))
initializer = helpers.build_initial_embedding_matrix(
vocab_dict, glove_dict, glove_vectors, hparams.embedding_dim)
elif hparams.vector_type == 'fastText':
# return load_embedding_vectors_fastText(vocab_array, hparams.glove_path, len(vocab_array))
fastText_vectors, fastText_dict = helpers.load_fastText_vectors(
hparams.fastText_path, vocab=set(vocab_array))
initializer = helpers.build_initial_embedding_matrix(
vocab_dict, fastText_dict, fastText_vectors, hparams.embedding_dim)
# if hparams.glove_path and hparams.vocab_path:
# tf.logging.info("Loading Glove embeddings...")
# #加载词汇表，训练数据中包含的
# vocab_array, vocab_dict = helpers.load_vocab(hparams.vocab_path)
# #
# glove_vectors, glove_dict = helpers.load_glove_vectors(hparams.glove_path, vocab=set(vocab_array))
#
# initializer = helpers.build_initial_embedding_matrix(vocab_dict, glove_dict, glove_vectors,
# hparams.embedding_dim)
else:
tf.logging.info(
"No glove/vocab path specificed, starting with random embeddings.")
initializer = tf.random_uniform_initializer(-0.25, 0.25) # 随机均匀
# If initializer is a constant, do not specify shape.
return tf.get_variable(
"word_embeddings",
# shape=[hparams.vocab_size, hparams.embedding_dim],
initializer=initializer)
def get_embeddings(hparams):
if hparams.glove_path and hparams.vocab_path:
tf.logging.info("Loading Glove embeddings...")
vocab_array, vocab_dict = helpers.load_vocab(hparams.vocab_path)
glove_vectors, glove_dict = helpers.load_glove_vectors(hparams.glove_path, vocab=set(vocab_array))
initializer = helpers.build_initial_embedding_matrix(vocab_dict, glove_dict, glove_vectors, hparams.embedding_dim)
else:
tf.logging.info("No glove/vocab path specificed, starting with random embeddings.") #当eval或者再次train的时候，dual encoder中的get_embedding仍然会触发这句话
initializer = tf.random_uniform_initializer(-0.25, 0.25) #一个initializer object,因而无需指定形状（因为get_variable时会先指定一个shape）？
return tf.get_variable(
"word_embeddings",
shape=[hparams.vocab_size, hparams.embedding_dim],
initializer=initializer)
def get_embeddings(hparams):
if hparams.glove_path and hparams.vocab_path:
tf.logging.info("Loading Glove embeddings...")
vocab_array, vocab_dict = helpers.load_vocab(hparams.vocab_path)
glove_vectors, glove_dict = helpers.load_glove_vectors(hparams.glove_path, vocab=set(vocab_array))
initializer = helpers.build_initial_embedding_matrix(vocab_dict, glove_dict, glove_vectors, hparams.embedding_dim)
else:
tf.logging.info("No glove/vocab path specificed, starting with random embeddings.")
initializer = tf.random_uniform_initializer(-0.25, 0.25)
return tf.get_variable(
"word_embeddings",
shape=[hparams.vocab_size, hparams.embedding_dim],
initializer=initializer)
def get_embeddings(hparams):
if hparams.w2v_path and hparams.vocab_path:
tf.logging.info("Loading Glove embeddings...")
vocab_array, vocab_dict = helpers.load_vocab(hparams.vocab_path)
w2v_vectors, w2v_dict = helpers.load_w2v_vectors(
hparams.w2v_path, vocab=set(vocab_array))
initializer = helpers.build_initial_embedding_matrix(
vocab_dict, w2v_dict, w2v_vectors, hparams.embedding_dim)
#return tf.Variable(initializer, name="vocab_w")
else:
tf.logging.info(
"No w2v/vocab path specificed, starting with random embeddings.")
initializer = tf.random_uniform_initializer(-0.25, 0.25)
return tf.get_variable("word_embeddings",
shape=[hparams.vocab_size, hparams.embedding_dim],
initializer=initializer)
def get_embeddings(hparams, glove):
if glove:
tf.logging.info("Loading GloVe embedding ...")
glove_path = "data/glove.6B.100d.txt"
vocab_path = "data/vocabulary.txt"
vocab_array, vocab_dict = helpers.load_vocab(vocab_path)
glove_vectors, glove_dict = helpers.load_glove_vectors(
glove_path, vocab=set(vocab_array))
initializer = helpers.build_initial_embedding_matrix(
vocab_dict, glove_dict, glove_vectors, hparams.embedding_dim)
else:
tf.logging.info("Loading random embedding ...")
initializer = tf.random_uniform_initializer(-0.25, 0.25)
return tf.get_variable("word_embeddings",
shape=[hparams.vocab_size, hparams.embedding_dim],
initializer=initializer)
