def test_get_embedding(self):
"""Tests :func:`~texar.modules.embedder.embedder_utils.get_embedding`.
"""
vocab_size = 100
emb = embedder_utils.get_embedding(num_embeds=vocab_size)
self.assertEqual(emb.shape[0], vocab_size)
self.assertEqual(emb.shape[1],
embedder_utils.default_embedding_hparams()["dim"])
hparams = {
"initializer": {
"type": "torch.nn.init.uniform_",
"kwargs": {
'a': -0.1,
'b': 0.1
}
}
}
emb = embedder_utils.get_embedding(
hparams=hparams,
num_embeds=vocab_size,
)
self.assertEqual(emb.shape[0], vocab_size)
self.assertEqual(emb.shape[1],
embedder_utils.default_embedding_hparams()["dim"])
def __init__(self, embedding=None, vocab_size=None, hparams=None):
ModuleBase.__init__(self, hparams)
self._vocab_size = vocab_size
self._embedding = None
self.sampling_method = self._hparams.sampling_method
with tf.variable_scope(self.variable_scope):
if self._hparams.initializer:
tf.get_variable_scope().set_initializer( \
layers.get_initializer(self._hparams.initializer))
if self._hparams.position_embedder.name == 'sinusoids':
self.position_embedder = \
position_embedders.SinusoidsSegmentalPositionEmbedder( \
self._hparams.position_embedder.hparams)
if self._hparams.use_embedding:
if embedding is None and vocab_size is None:
raise ValueError("""If 'embedding' is not provided,
'vocab_size' must be specified.""")
if isinstance(embedding, (tf.Tensor, tf.Variable)):
self._embedding = embedding
else:
self._embedding = embedder_utils.get_embedding(
self._hparams.embedding,
embedding,
vocab_size,
variable_scope=self.variable_scope)
self._embed_dim = shape_list(self._embedding)[-1]
if self._hparams.zero_pad:
self._embedding = tf.concat( \
(tf.zeros(shape=[1, self._embed_dim]),\
self._embedding[1:, :]), 0)
if self._vocab_size is None:
self._vocab_size = self._embedding.get_shape().as_list()[0]
self.output_layer = \
self.build_output_layer(shape_list(self._embedding)[-1])
def _init_parameterized_embedding(self, init_value, num_embeds, hparams):
self._embedding = embedder_utils.get_embedding(
hparams, init_value, num_embeds, self.variable_scope)
if hparams.trainable:
self._add_trainable_variable(self._embedding)
self._num_embeds = self._embedding.get_shape().as_list()[0]
self._dim = self._embedding.get_shape().as_list()[1:]
self._dim_rank = len(self._dim)
if self._dim_rank == 1:
self._dim = self._dim[0]
def __init__(self, num_embeds=None, init_value=None, hparams=None):
ModuleBase.__init__(self, hparams)
if num_embeds is not None or init_value is not None:
self._embedding = Parameter(
embedder_utils.get_embedding(num_embeds, init_value, hparams))
self._num_embeds = self._embedding.shape[0]
self._dim = self._embedding.shape[1:]
self._dim_rank = len(self._dim)
if self._dim_rank == 1:
self._dim = self._dim[0]
def test_get_embedding(self):
"""Tests :func:`~texar.modules.embedder.embedder_utils.get_embedding`.
"""
vocab_size = 100
emb = embedder_utils.get_embedding(num_embeds=vocab_size)
self.assertEqual(emb.shape[0].value, vocab_size)
self.assertEqual(emb.shape[1].value,
embedder_utils.default_embedding_hparams()["dim"])
hparams = {
"initializer": {
"type": tf.random_uniform_initializer(minval=-0.1, maxval=0.1)
},
"regularizer": {
"type": tf.keras.regularizers.L1L2(0.1, 0.1)
}
}
emb = embedder_utils.get_embedding(hparams=hparams,
num_embeds=vocab_size,
variable_scope='embedding_2')
self.assertEqual(emb.shape[0].value, vocab_size)
self.assertEqual(emb.shape[1].value,
embedder_utils.default_embedding_hparams()["dim"])
def __init__(self, num_embeds: Optional[int] = None,
init_value: Optional[torch.Tensor] = None, hparams=None):
super().__init__(hparams)
if num_embeds is not None or init_value is not None:
self._embedding = nn.Parameter(embedder_utils.get_embedding(
num_embeds, init_value, hparams))
self._num_embeds = self._embedding.size(0)
self._dim_rank = self._embedding.dim() - 1
if self._dim_rank == 1:
self._dim = self._embedding.size(1)
else:
self._dim = self._embedding.size()[1:] # type: ignore