def testBowEncoderSparseTensor(self):
with self.cached_session() as sess:
docs = [[0, 1], [2, 3]]
sparse_docs = sparse_ops.dense_to_sparse_tensor(docs)
enc = encoders.bow_encoder(sparse_docs, 4, 3)
sess.run(variables.global_variables_initializer())
self.assertAllEqual([2, 3], enc.eval().shape)
def bow_encoder(ids,
vocab_size,
embed_dim,
sparse_lookup=True,
initializer=None,
regularizer=None,
trainable=True,
scope=None,
reuse=None):
"""Maps a sequence of symbols to a vector per example by averaging embeddings.
Args:
ids: `[batch_size, doc_length]` `Tensor` or `SparseTensor` of type
`int32` or `int64` with symbol ids.
vocab_size: Integer number of symbols in vocabulary.
embed_dim: Integer number of dimensions for embedding matrix.
sparse_lookup: `bool`, if `True`, converts ids to a `SparseTensor`
and performs a sparse embedding lookup. This is usually faster,
but not desirable if padding tokens should have an embedding. Empty rows
are assigned a special embedding.
initializer: An initializer for the embeddings, if `None` default for
current scope is used.
regularizer: Optional regularizer for the embeddings.
trainable: If `True` also add variables to the graph collection
`GraphKeys.TRAINABLE_VARIABLES` (see tf.Variable).
scope: Optional string specifying the variable scope for the op, required
if `reuse=True`.
reuse: If `True`, variables inside the op will be reused.
Returns:
Encoding `Tensor` `[batch_size, embed_dim]` produced by
averaging embeddings.
Raises:
ValueError: If `embed_dim` or `vocab_size` are not specified.
"""
if not vocab_size or not embed_dim:
raise ValueError('Must specify vocab size and embedding dimension')
with variable_scope.variable_scope(scope,
'bow_encoder', [ids],
reuse=reuse):
embeddings = variables.model_variable('embeddings',
shape=[vocab_size, embed_dim],
initializer=initializer,
regularizer=regularizer,
trainable=trainable)
if sparse_lookup:
if isinstance(ids, sparse_tensor.SparseTensor):
sparse_ids = ids
else:
sparse_ids = sparse_ops.dense_to_sparse_tensor(ids)
return contrib_embedding_ops.safe_embedding_lookup_sparse(
[embeddings], sparse_ids, combiner='mean', default_id=0)
else:
if isinstance(ids, sparse_tensor.SparseTensor):
raise TypeError('ids are expected to be dense Tensor, got: %s',
ids)
return math_ops.reduce_mean(embedding_ops.embedding_lookup(
embeddings, ids),
axis=1)
def test_dense_to_sparse_tensor_2d(self):
with self.cached_session() as sess:
st = sparse_ops.dense_to_sparse_tensor([[1, 2, 0, 0], [3, 4, 5, 0]])
result = sess.run(st)
self.assertAllEqual([[0, 0], [0, 1], [1, 0], [1, 1], [1, 2]],
result.indices)
self.assertAllEqual([1, 2, 3, 4, 5], result.values)
self.assertAllEqual([2, 4], result.dense_shape)
def test_dense_to_sparse_tensor_unknown_1d_shape(self):
with self.cached_session() as sess:
tensor = array_ops.placeholder(shape=[None], dtype=dtypes.int32)
st = sparse_ops.dense_to_sparse_tensor(tensor)
result = sess.run(st, feed_dict={tensor: [0, 100, 0, 3]})
self.assertAllEqual([[1], [3]], result.indices)
self.assertAllEqual([100, 3], result.values)
self.assertAllEqual([4], result.dense_shape)
def test_dense_to_sparse_unknown_rank(self):
ph = array_ops.placeholder(dtype=dtypes.int32)
with self.cached_session() as sess:
st = sparse_ops.dense_to_sparse_tensor(ph)
result = sess.run(st, feed_dict={ph: [[1, 2, 0, 0], [3, 4, 5, 0]]})
self.assertAllEqual([[0, 0], [0, 1], [1, 0], [1, 1], [1, 2]],
result.indices)
self.assertAllEqual([1, 2, 3, 4, 5], result.values)
self.assertAllEqual([2, 4], result.dense_shape)
def test_dense_to_sparse_tensor_1d_str(self):
with self.cached_session() as sess:
st = sparse_ops.dense_to_sparse_tensor([b'qwe', b'', b'ewq', b''])
result = sess.run(st)
self.assertEqual(result.indices.dtype, np.int64)
self.assertEqual(result.values.dtype, np.object)
self.assertEqual(result.dense_shape.dtype, np.int64)
self.assertAllEqual([[0], [2]], result.indices)
self.assertAllEqual([b'qwe', b'ewq'], result.values)
self.assertAllEqual([4], result.dense_shape)
def test_dense_to_sparse_tensor_1d_bool(self):
with self.cached_session() as sess:
st = sparse_ops.dense_to_sparse_tensor([True, False, True, False])
result = sess.run(st)
self.assertEqual(result.indices.dtype, np.int64)
self.assertEqual(result.values.dtype, np.bool)
self.assertEqual(result.dense_shape.dtype, np.int64)
self.assertAllEqual([[0], [2]], result.indices)
self.assertAllEqual([True, True], result.values)
self.assertAllEqual([4], result.dense_shape)
def test_dense_to_sparse_tensor_1d_float(self):
with self.cached_session() as sess:
st = sparse_ops.dense_to_sparse_tensor([1.5, 0.0, 2.3, 0.0])
result = sess.run(st)
self.assertEqual(result.indices.dtype, np.int64)
self.assertEqual(result.values.dtype, np.float32)
self.assertEqual(result.dense_shape.dtype, np.int64)
self.assertAllEqual([[0], [2]], result.indices)
self.assertAllClose([1.5, 2.3], result.values)
self.assertAllEqual([4], result.dense_shape)
def test_dense_to_sparse_tensor_1d(self):
with self.cached_session() as sess:
st = sparse_ops.dense_to_sparse_tensor([1, 0, 2, 0])
result = sess.run(st)
self.assertEqual(result.indices.dtype, np.int64)
self.assertEqual(result.values.dtype, np.int32)
self.assertEqual(result.dense_shape.dtype, np.int64)
self.assertAllEqual([[0], [2]], result.indices)
self.assertAllEqual([1, 2], result.values)
self.assertAllEqual([4], result.dense_shape)
def test_dense_to_sparse_tensor_unknown_3d_shape(self):
with self.cached_session() as sess:
tensor = array_ops.placeholder(
shape=[None, None, None], dtype=dtypes.int32)
st = sparse_ops.dense_to_sparse_tensor(tensor)
result = sess.run(st,
feed_dict={
tensor: [[[1, 2, 0, 0], [3, 4, 5, 0]],
[[7, 8, 0, 0], [9, 0, 0, 0]]]
})
self.assertAllEqual([[0, 0, 0], [0, 0, 1], [0, 1, 0], [0, 1, 1], [0, 1, 2],
[1, 0, 0], [1, 0, 1], [1, 1, 0]], result.indices)
self.assertAllEqual([1, 2, 3, 4, 5, 7, 8, 9], result.values)
self.assertAllEqual([2, 2, 4], result.dense_shape)
def testBowEncoderSparseTensorDenseLookup(self):
with self.cached_session():
docs = [[0, 1]]
sparse_docs = sparse_ops.dense_to_sparse_tensor(docs)
with self.assertRaises(TypeError):
encoders.bow_encoder(sparse_docs, 4, 3, sparse_lookup=False)
