def from_config(cls, config):
return cls(
inputs=[[
preprocessors_module.deserialize(preprocessor)
for preprocessor in preprocessors
] for preprocessors in config["inputs"]],
outputs=[[
preprocessors_module.deserialize(preprocessor)
for preprocessor in preprocessors
] for preprocessors in config["outputs"]],
)
def test_multi_label_deserialize_without_error():
encoder = encoders.MultiLabelEncoder()
dataset = tf.data.Dataset.from_tensor_slices([1, 2]).batch(32)
encoder = preprocessors.deserialize(preprocessors.serialize(encoder))
assert encoder.transform(dataset) is dataset
def test_softmax_deserialize_without_error():
postprocessor = postprocessors.SoftmaxPostprocessor()
dataset = tf.data.Dataset.from_tensor_slices([1, 2]).batch(32)
postprocessor = preprocessors.deserialize(
preprocessors.serialize(postprocessor))
assert postprocessor.transform(dataset) is dataset
def test_one_hot_encoder_deserialize_transforms_to_np():
encoder = encoders.OneHotEncoder(["a", "b", "c"])
encoder.fit(np.array(["a", "b", "a"]))
encoder = preprocessors.deserialize(preprocessors.serialize(encoder))
one_hot = encoder.transform(
tf.data.Dataset.from_tensor_slices([["a"], ["c"], ["b"]]).batch(2))
for data in one_hot:
assert data.shape[1:] == [3]
def from_config(cls, config):
init_config = {
"column_types": config["column_types"],
"column_names": config["column_names"],
}
obj = cls(**init_config)
obj.layer = preprocessors.deserialize(config["encoding_layer"])
for encoding_layer, vocab in zip(obj.layer.encoding_layers,
config["encoding_vocab"]):
if encoding_layer is not None:
encoding_layer.set_vocabulary(vocab)
return obj
def from_config(cls, config):
config["preprocessor"] = preprocessors.deserialize(
config["preprocessor"])
return super().from_config(config)
