random_augmentation_pipeline

RandomAugmentation1DPipeline(
    layers: list[keras.Layer],
    augmentations_per_sample: int = 1,
    rate: float = 1.0,
    batchwise: bool = True,
    force_training: bool = False,
    **kwargs={},
)

Apply repeated random layer choices to an entire batch.

Each round samples one layer for the batch, with replacement, and applies it with probability rate. Inference is unchanged unless force_training is enabled. Candidate layers must produce compatible shapes and structures.

Base class: RandomChoice.

Inherited from RandomChoice: build(), from_config().

Inherited from BaseAugmentation: augment_masks(), augment_sample(), augment_samples(), augment_targets(), get_random_transformations().

Parameters

Parameters of RandomAugmentation1DPipeline
Name	Type	Default	Description
`layers`	`list[keras.Layer]`	Required	Nonempty list of candidate augmentation layers.
`augmentations_per_sample`	`int`	1	Number of batchwise selection rounds. Zero leaves the input unchanged.
`rate`	`float`	1.0	Probability of applying each round, between zero and one.
`batchwise`	`bool`	True	Must be True; per-example layer selection is unsupported.
`force_training`	`bool`	False	Apply augmentation even when training is False or None.
`**kwargs`	`Any`	{}	Base augmentation options, including seed and data_format.

Raises

Errors raised by RandomAugmentation1DPipeline
Type	Description
`ValueError`	The layer list is empty, the round count is not a nonnegative integer, rate is outside [0, 1], or explicit transformations are used.
`NotImplementedError`	batchwise is False.

augmentations_per_sample = augmentations_per_sample

rate = rate

force_training = force_training

batch_augment(inputs, transformations=None)

call(inputs, training=None, transformations=None)

get_config()

RandomAugmentation2DPipeline(
    layers: list[keras.Layer],
    augmentations_per_sample: int = 1,
    rate: float = 1.0,
    batchwise: bool = True,
    force_training: bool = False,
    **kwargs={},
)

The same batchwise composition contract for image transforms.

Base class: RandomAugmentation1DPipeline.

Inherited from RandomAugmentation1DPipeline: batch_augment(), call(), get_config().

Inherited from RandomChoice: build(), from_config().

Inherited from BaseAugmentation: augment_masks(), augment_sample(), augment_samples(), augment_targets(), get_random_transformations().