def _testTokensToChars(self, tokens, expected_chars, expected_lengths):
expected_chars = tf.nest.map_structure(tf.compat.as_bytes,
expected_chars)
chars = text.tokens_to_chars(tf.constant(tokens, dtype=tf.string))
chars, lengths = chars.to_tensor(), chars.row_lengths()
chars, lengths = self.evaluate([chars, lengths])
self.assertListEqual(expected_chars, chars.tolist())
self.assertListEqual(expected_lengths, lengths.tolist())
def make_features(self, element=None, features=None, training=None):
"""Converts words to characters."""
if features is None:
features = {}
if "char_ids" in features:
return features
if "chars" in features:
chars = features["chars"]
else:
features = super(CharEmbedder, self).make_features(
element=element, features=features, training=training)
chars = text.tokens_to_chars(features["tokens"])
chars = chars.to_tensor(default_value=constants.PADDING_TOKEN)
features["char_ids"] = self.tokens_to_ids.lookup(chars)
return features
def _testTokensToChars(self, tokens, expected_chars):
expected_chars = tf.nest.map_structure(tf.compat.as_bytes,
expected_chars)
chars = text.tokens_to_chars(tf.constant(tokens, dtype=tf.string))
self.assertListEqual(chars.to_list(), expected_chars)
