def char_lm(key, sentence, labels):
return ar_lm(key,
sentence,
labels,
char_embeddings=ex.embeddings(
id_space_size=len(get_chars()),
embedding_size=qnd.FLAGS.char_embedding_size,
name='char_embeddings'))
def word2sent2doc(document, *, word_space_size, word_embedding_size,
**rd2sent2doc_hyperparams):
assert ex.static_rank(document) == 3
with tf.variable_scope("word_embeddings"):
word_embeddings = tf.gather(
ex.embeddings(id_space_size=word_space_size,
embedding_size=word_embedding_size,
name="word_embeddings"), ex.flatten(document))
return rd2sent2doc(document,
word_embeddings,
save_memory=True,
**rd2sent2doc_hyperparams)
def word2sent2doc(document,
*,
word_space_size,
word_embedding_size,
**rd2sent2doc_hyperparams):
assert ex.static_rank(document) == 3
with tf.variable_scope("word_embeddings"):
word_embeddings = tf.gather(
ex.embeddings(id_space_size=word_space_size,
embedding_size=word_embedding_size,
name="word_embeddings"),
ex.flatten(document))
return rd2sent2doc(document,
word_embeddings,
save_memory=True,
**rd2sent2doc_hyperparams)
def char2word2sent2doc(document, *, words, char_space_size,
char_embedding_size, **ar2word2sent2doc_hyperparams):
"""
The argument `document` is in the shape of
(#examples, #sentences per document, #words per sentence).
"""
assert ex.static_rank(document) == 3
assert ex.static_rank(words) == 2
with tf.variable_scope("char_embeddings"):
char_embeddings = ex.embeddings(id_space_size=char_space_size,
embedding_size=char_embedding_size,
name="char_embeddings")
return ar2word2sent2doc(document,
words=words,
char_embeddings=char_embeddings,
**ar2word2sent2doc_hyperparams)
def char2word2sent2doc(document,
*,
words,
char_space_size,
char_embedding_size,
**ar2word2sent2doc_hyperparams):
"""
The argument `document` is in the shape of
(#examples, #sentences per document, #words per sentence).
"""
assert ex.static_rank(document) == 3
assert ex.static_rank(words) == 2
with tf.variable_scope("char_embeddings"):
char_embeddings = ex.embeddings(id_space_size=char_space_size,
embedding_size=char_embedding_size,
name="char_embeddings")
return ar2word2sent2doc(document,
words=words,
char_embeddings=char_embeddings,
**ar2word2sent2doc_hyperparams)