function
interval_mask
PythonSample one interval per example; returned mask broadcasts over channels.
interval_mask(layer, shape, axis, minimum, maximum)Sample one interval per example; returned mask broadcasts over channels.
random_cutout
Signal-only occlusion with explicit half-open intervals and owned RNG.
interval_maskfunctionSample one interval per example; returned mask broadcasts over channels.RandomCutout1DclassReplace random temporal intervals in each example during training.RandomCutout2DclassReplace random rectangular regions in each example during training.
class
RandomCutout1D
PythonReplace random temporal intervals in each example during training.
RandomCutout1D(factor=0.1, cutouts=1, fill_mode='constant', fill_value=0.0, **kwargs={})Replace random temporal intervals in each example during training.
Regions share their positions across channels. Targets and masks remain unchanged; only signal values are occluded. Overlapping regions are allowed.
Base class: BaseAugmentation1D.
Inherited from BaseAugmentation: augment_masks(), augment_sample(), augment_targets(), batch_augment(), call().
Parameters
| Name | Type | Default | Description |
|---|---|---|---|
factor | float | tuple[float, float] | 0.1 | Fractional size bounds in [0, 1] for each spatial axis. A scalar sets the upper bound with a zero lower bound; a pair sets both bounds. |
cutouts | int | 1 | Nonnegative number of regions sampled per example. |
fill_mode | str | 'constant' | "constant" for a fixed value or "normal" for Gaussian noise. |
fill_value | float | 0.0 | Constant fill value, or the nonnegative standard deviation of zero-mean Gaussian noise when fill_mode is "normal". |
**kwargs | Any | {} | Base augmentation options, including seed and data_format. |
Raises
| Type | Description |
|---|---|
ValueError | Size bounds, region count, fill mode or noise deviation violate these constraints. |
method
get_random_transformations(input_shape)method
augment_samples
Pythonaugment_samples(inputs)method
get_config
Pythonget_config()class
RandomCutout2D
PythonReplace random rectangular regions in each example during training.
RandomCutout2D(factor=0.1, cutouts=1, fill_mode='constant', fill_value=0.0, **kwargs={})Replace random rectangular regions in each example during training.
Regions share their positions across channels. Targets and masks remain unchanged; only signal values are occluded. Overlapping regions are allowed.
Base class: BaseAugmentation2D.
Inherited from BaseAugmentation: augment_masks(), augment_sample(), augment_targets(), batch_augment(), call().
Parameters
| Name | Type | Default | Description |
|---|---|---|---|
factor | float | tuple[float, float] | 0.1 | Fractional size bounds in [0, 1] for each spatial axis. A scalar sets the upper bound with a zero lower bound; a pair sets both bounds. |
cutouts | int | 1 | Nonnegative number of regions sampled per example. |
fill_mode | str | 'constant' | "constant" for a fixed value or "normal" for Gaussian noise. |
fill_value | float | 0.0 | Constant fill value, or the nonnegative standard deviation of zero-mean Gaussian noise when fill_mode is "normal". |
**kwargs | Any | {} | Base augmentation options, including seed and data_format. |
Raises
| Type | Description |
|---|---|
ValueError | Size bounds, region count, fill mode or noise deviation violate these constraints. |
method
get_random_transformations(input_shape)method
augment_samples
Pythonaugment_samples(inputs)method
get_config
Pythonget_config()