def test_output_values_with_sampled_candidates(self):
"""Verifies the values for given sampled_candidates."""
with self.test_session():
sp_values = sparse_tensor_lib.SparseTensor(
values=["a", "a", "b", "c", "d", "e", "f"],
indices=[[1, 0], [2, 0], [2, 1], [2, 2], [2, 3], [2, 4],
[2, 5]],
dense_shape=[3, 6])
params = constant_op.constant([.1, .2, .3])
sampled_candidates = [[1, 0], [2, 1], [3, 2]]
sampled_result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params,
sp_values,
sampled_candidates=constant_op.constant(sampled_candidates),
hash_key=self._hash_key)
full_result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, dimension=4, hash_key=self._hash_key)
sampled_result_val = sampled_result.eval()
full_result_val = full_result.eval()
self.assertEqual(sampled_result_val.shape, (3, 2))
for i in range(len(sampled_candidates)):
self.assertAllClose(sampled_result_val[i],
full_result_val[i, sampled_candidates[i]])
def test_distributive_property(self):
"""Verifies the distributive property of matrix multiplication."""
with self.test_session():
params = constant_op.constant([.1, .2, .3])
sp_values_a = sparse_tensor_lib.SparseTensor(values=["a"],
indices=[[0, 0]],
dense_shape=[3, 1])
sp_values_b = sparse_tensor_lib.SparseTensor(values=["b"],
indices=[[2, 0]],
dense_shape=[3, 1])
sp_values_c = sparse_tensor_lib.SparseTensor(values=["c"],
indices=[[2, 0]],
dense_shape=[3, 1])
sp_values = sparse_tensor_lib.SparseTensor(values=["a", "b", "c"],
indices=[[0, 0], [2, 0],
[2, 1]],
dense_shape=[3, 2])
result_a = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values_a, dimension=4, hash_key=self._hash_key)
result_b = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values_b, dimension=4, hash_key=self._hash_key)
result_c = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values_c, dimension=4, hash_key=self._hash_key)
result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, dimension=4, hash_key=self._hash_key)
result_abc = math_ops.add_n([result_a, result_b, result_c])
self.assertAllClose(result.eval(), result_abc.eval())
def test_distributive_property(self):
"""Verifies the distributive property of matrix multiplication."""
with self.cached_session():
params = constant_op.constant([.1, .2, .3])
sp_values_a = sparse_tensor_lib.SparseTensor(
values=["a"], indices=[[0, 0]], dense_shape=[3, 1])
sp_values_b = sparse_tensor_lib.SparseTensor(
values=["b"], indices=[[2, 0]], dense_shape=[3, 1])
sp_values_c = sparse_tensor_lib.SparseTensor(
values=["c"], indices=[[2, 0]], dense_shape=[3, 1])
sp_values = sparse_tensor_lib.SparseTensor(
values=["a", "b", "c"],
indices=[[0, 0], [2, 0], [2, 1]],
dense_shape=[3, 2])
result_a = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values_a, dimension=4, hash_key=self._hash_key)
result_b = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values_b, dimension=4, hash_key=self._hash_key)
result_c = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values_c, dimension=4, hash_key=self._hash_key)
result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, dimension=4, hash_key=self._hash_key)
result_abc = math_ops.add_n([result_a, result_b, result_c])
self.assertAllClose(result.eval(), result_abc.eval())
def test_output_values(self):
"""Verifies the values in a trivial case."""
with self.cached_session():
sp_values = sparse_tensor_lib.SparseTensor(
values=["a"], indices=[[1, 0]], dense_shape=[3, 1])
params = constant_op.constant([.1, .2, .3])
result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, dimension=5, hash_key=self._hash_key)
self.assertAllClose(result.eval(), [[0., 0., 0., 0.,
0.], [.3, .2, .2, .3, .1],
[0., 0., 0., 0., 0.]])
def test_output_shape(self):
"""Verifies the shape of the output tensor."""
with self.cached_session():
sp_values = sparse_tensor_lib.SparseTensor(
values=["a", "a", "b", "c", "d", "e", "f"],
indices=[[1, 0], [2, 0], [2, 1], [2, 2], [2, 3], [2, 4], [2, 5]],
dense_shape=[3, 6])
params = constant_op.constant([.1, .2, .3])
result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, dimension=4, hash_key=self._hash_key)
self.assertEqual(result.eval().shape, (3, 4))
def test_output_values(self):
"""Verifies the values in a trivial case."""
with self.test_session():
sp_values = sparse_tensor_lib.SparseTensor(
values=["a"], indices=[[1, 0]], dense_shape=[3, 1])
params = constant_op.constant([.1, .2, .3])
result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, dimension=5, hash_key=self._hash_key)
self.assertAllClose(result.eval(), [[0., 0., 0., 0., 0.],
[.3, .2, .2, .3, .1],
[0., 0., 0., 0., 0.]])
def test_output_shape(self):
"""Verifies the shape of the output tensor."""
with self.test_session():
sp_values = sparse_tensor_lib.SparseTensor(
values=["a", "a", "b", "c", "d", "e", "f"],
indices=[[1, 0], [2, 0], [2, 1], [2, 2], [2, 3], [2, 4], [2, 5]],
dense_shape=[3, 6])
params = constant_op.constant([.1, .2, .3])
result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, dimension=4, hash_key=self._hash_key)
self.assertEqual(result.eval().shape, (3, 4))
def test_output_values_with_sampled_candidates(self):
"""Verifies the values for given sampled_candidates."""
with self.test_session():
sp_values = tf.SparseTensor(
values=["a", "a", "b", "c", "d", "e", "f"],
indices=[[1, 0], [2, 0], [2, 1], [2, 2], [2, 3], [2, 4], [2, 5]],
dense_shape=[3, 6])
params = tf.constant([.1, .2, .3])
sampled_candidates = [[1, 0], [2, 1], [3, 2]]
sampled_result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, sampled_candidates=tf.constant(sampled_candidates),
hash_key=self._hash_key)
full_result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, dimension=4, hash_key=self._hash_key)
sampled_result_val = sampled_result.eval()
full_result_val = full_result.eval()
self.assertEqual(sampled_result_val.shape, (3, 2))
for i in range(len(sampled_candidates)):
self.assertAllClose(sampled_result_val[i],
full_result_val[i, sampled_candidates[i]])
def test_output_values_with_sign_hash(self):
"""Verifies the values in a trivial case with hash_signs=True."""
with self.test_session():
sp_values = tf.SparseTensor(
values=["a"], indices=[[1, 0]], dense_shape=[3, 1])
params = tf.constant([.1, .1, .1])
result = embedding_ops._sampled_scattered_embedding_lookup_sparse(
params, sp_values, dimension=4, with_sign_hash=True,
hash_key=self._hash_key)
self.assertAllClose(result.eval(),
[[0., 0., 0., 0.],
[-.1, -.1, -.1, .1],
[0., 0., 0., 0.]])
