def test_serialize_deserialize_tensor_value(self): x = tf.constant(10.0).numpy() value_proto, value_type = executor_service_utils.serialize_tensor_value(x) self.assertIsInstance(value_proto, executor_pb2.Value) self.assertEqual(str(value_type), 'float32') y, type_spec = executor_service_utils.deserialize_tensor_value(value_proto) self.assertEqual(str(type_spec), 'float32') self.assertTrue(np.array_equal(x, y))
def test_serialize_deserialize_tensor_value_with_nontrivial_shape(self):
x = tf.constant([10, 20, 30]).numpy()
value_proto, value_type = executor_service_utils.serialize_tensor_value(
x, (tf.int32, [3]))
self.assertIsInstance(value_proto, executor_pb2.Value)
self.assertEqual(str(value_type), 'int32[3]')
y, type_spec = executor_service_utils.deserialize_tensor_value(value_proto)
self.assertEqual(str(type_spec), 'int32[3]')
self.assertTrue(np.array_equal(x, y))
def test_serialize_deserialize_tensor_value_with_different_dtype(self):
x = tf.constant(10.0).numpy()
value_proto, value_type = (
executor_service_utils.serialize_tensor_value(x, tf.int32))
self.assertIsInstance(value_proto, executor_pb2.Value)
self.assertEqual(str(value_type), 'int32')
y, type_spec = executor_service_utils.deserialize_tensor_value(value_proto)
self.assertEqual(str(type_spec), 'int32')
self.assertEqual(y, 10)
def test_serialize_deserialize_tensor_value_with_bad_shape(self):
x = tf.constant([10, 20, 30]).numpy()
with self.assertRaises(TypeError):
executor_service_utils.serialize_tensor_value(x, tf.int32)
