def test_maybe_tile_tensor_for_mc_tiles_scalar_tensor(self):
num_samples = 16
x = 3.14
tensor_in = tf.constant(x, dtype=tf.float32)
tensor_out = monte_carlo_manager._maybe_tile_tensor_for_mc(
tensor_in, num_samples)
self.assertEqual(tensor_out.shape.as_list(), [num_samples])
with self.session() as session:
tensor_out_eval = session.run(tensor_out)
self.assertAllEqual(tensor_out_eval,
np.asarray([x] * num_samples, dtype=np.float32))
def test_maybe_tile_tensor_for_mc_does_not_tile_2d_tensor(self):
np.random.seed(0)
num_samples = 16
num_dims = 7
x = np.random.uniform(size=[num_samples, num_dims])
tensor_in = tf.constant(x)
tensor_out = monte_carlo_manager._maybe_tile_tensor_for_mc(
tensor_in, num_samples)
self.assertAllEqual(tensor_out.shape.as_list(), [num_samples, num_dims])
with self.session() as session:
tensor_out_eval = session.run(tensor_out)
self.assertAllEqual(tensor_out_eval, x)
def test_reshape_initial_state_for_mc_reshapes_tuple(self):
np.random.seed(0)
num_samples = 16
tensors_in = (tf.random_uniform(shape=[3]), tf.random_uniform(shape=[7]),
tf.random_uniform(shape=()))
tensors_out = monte_carlo_manager._reshape_initial_state_for_mc(
tensors_in, num_samples)
expected_tensors_out = (monte_carlo_manager._maybe_tile_tensor_for_mc(
tensors_in[0], num_samples),
monte_carlo_manager._maybe_tile_tensor_for_mc(
tensors_in[1], num_samples),
monte_carlo_manager._maybe_tile_tensor_for_mc(
tensors_in[2], num_samples))
for i in range(len(expected_tensors_out)):
self.assertAllEqual(tensors_out[i].shape.as_list(),
expected_tensors_out[i].shape.as_list())
with self.session() as session:
(tensors_out_eval, expected_tensors_out_eval) = session.run(
(tensors_out, expected_tensors_out))
for i in range(len(expected_tensors_out)):
self.assertAllEqual(tensors_out_eval[i], expected_tensors_out_eval[i])
