def test_input_requirement():
hp = hp_module.HyperParameters()
with pytest.raises(ValueError, match=r".*must specify.*"):
hm = aug_module.HyperImageAugment()
hm = aug_module.HyperImageAugment(input_shape=(None, None, 3))
model = hm.build(hp)
assert model.built
hm = aug_module.HyperImageAugment(input_tensor=keras.Input(shape=(32, 32,
3)))
model = hm.build(hp)
assert model.built
def test_model_construction_factor_zero():
hp = hp_module.HyperParameters()
hm = aug_module.HyperImageAugment(input_shape=(None, None, 3))
model = hm.build(hp)
# augment_layers search default space [0, 4], with default zero.
assert len(model.layers) == 1
hp = hp_module.HyperParameters()
hm = aug_module.HyperImageAugment(input_shape=(None, None, 3),
augment_layers=0)
model = hm.build(hp)
# factors default all zero, the model should only have input layer
assert len(model.layers) == 1
def test_hyperparameter_existence_and_hp_defaults_rand_aug():
hp = hp_module.HyperParameters()
hm = aug_module.HyperImageAugment(input_shape=(32, 32, 3),
augment_layers=[2, 5],
contrast=False)
hm.build(hp)
assert hp.get("augment_layers") == 2
def test_tf_version_too_low_error():
pp_module = aug_module.preprocessing
aug_module.preprocessing = None
with pytest.raises(ImportError, match="HyperImageAugment requires"):
aug_module.HyperImageAugment()
aug_module.preprocessing = pp_module
def test_hyperparameter_override_rand_aug():
hp = hp_module.HyperParameters()
hp.Fixed("randaug_mag", 1.0)
hp.Choice("randaug_count", [4])
hm = aug_module.HyperImageAugment(input_shape=(32, 32, 3),
augment_layers=[2, 4])
hm.build(hp)
assert hp.get("randaug_mag") == 1.0
assert hp.get("randaug_count") == 4
def test_transforms_search_space():
hm = aug_module.HyperImageAugment(input_shape=(32, 32, 3))
# Default choice
assert hm.transforms == [
("rotate", (0, 0.5)),
("translate_x", (0, 0.4)),
("translate_y", (0, 0.4)),
("contrast", (0, 0.3)),
]
hm = aug_module.HyperImageAugment(input_shape=(32, 32, 3),
rotate=0.3,
translate_x=[0.1, 0.5],
contrast=None)
assert hm.transforms == [
("rotate", (0, 0.3)),
("translate_x", (0.1, 0.5)),
("translate_y", (0, 0.4)),
]
def test_hyperparameter_override_fixed_aug():
hp = hp_module.HyperParameters()
hp.Fixed("factor_rotate", 0.9)
hp.Choice("factor_translate_x", [0.8])
hm = aug_module.HyperImageAugment(input_shape=(32, 32, 3),
augment_layers=0)
hm.build(hp)
assert hp.get("factor_rotate") == 0.9
assert hp.get("factor_translate_x") == 0.8
assert hp.get("factor_translate_y") == 0.0
assert hp.get("factor_contrast") == 0.0
def test_hyperparameter_selection_and_hp_defaults_fixed_aug():
hp = hp_module.HyperParameters()
hm = aug_module.HyperImageAugment(
input_shape=(32, 32, 3),
translate_x=[0.2, 0.4],
contrast=None,
augment_layers=0,
)
hm.build(hp)
# default value of default search space are always minimum.
assert hp.get("factor_rotate") == 0
assert hp.get("factor_translate_x") == 0.2
assert hp.get("factor_translate_y") == 0
assert "factor_contrast" not in hp.values
def test_model_construction_rand_aug():
hp = hp_module.HyperParameters()
hm = aug_module.HyperImageAugment(input_shape=(None, None, 3),
rotate=[0.2, 0.5])
model = hm.build(hp)
assert model.layers
assert model.name == "image_rand_augment"
# Output shape includes batch dimension.
assert model.output_shape == (None, None, None, 3)
out = model.predict(np.ones((1, 32, 32, 3)))
assert out.shape == (1, 32, 32, 3)
# Augment does not distort image when inferencing.
assert (out != 1).sum() == 0
