Skip to content

Documentation for AugmentationWithRandomSelection

Functionality

This class provides a base for applying a raw augmentation and selecting a random subset of the generated outputs. It inherits from AugmentationInterface to ensure a standard augmentation API.

Parameters

  • selection_size: Determines the number or proportion of results to select. If a float, it represents the fraction of augmented objects. If an int, it specifies the exact number of objects.

Motivation

The class enables flexible augmentation by allowing random selection from a set of generated transformations, useful for randomized testing and data diversification.

Inheritance

  • Inherits from AugmentationInterface.

Usage Example

Subclass AugmentationWithRandomSelection and implement the _raw_transform method. Use transform() to apply the augmentation and retrieve a random subset of results.

Documentation for AugmentationWithRandomSelection._raw_transform

Functionality

This method applies a raw transformation to the input object. It is abstract and must be implemented by subclasses to define custom augmentation logic. The method should return a list of augmented objects.

Parameters

  • object: The input data to be augmented. It can be of any type.

Usage

  • Purpose: To establish a blueprint for creating augmentations. Implement this method to generate initial augmented objects before any selection mechanism.

Example

Below is an example implementation: 

def _raw_transform(self, object: Any) -> List[Any]:
    return [f"Example 1: {object}",
            f"Example 2: {object}",
            f"Example 3: {object}"]

Documentation for AugmentationWithRandomSelection._select_augmentations

Functionality

Selects a subset of augmented objects based on the selection_size parameter provided at initialization. If selection_size is 1 or greater than or equal to the total number of augmentations, the full augmented list is returned. For values less than 1, it randomly selects a proportion of augmented objects. For integer values other than 1, it selects the specified number of random samples.

Parameters

  • transformed (List[Any]): A list of augmented objects obtained from _raw_transform.

Usage

  • Purpose: To filter and randomly choose a subset of augmentations after generating all possible augmentations.

Example

If _raw_transform returns ["a", "b", "c", "d"] and selection_size is 0.5, this method will randomly select about 2 augmentations.

Documentation for AugmentationWithRandomSelection.transform

Functionality

Transforms the input object by first applying the raw augmentation via _raw_transform and then randomly selecting a subset with _select_augmentations.

Parameters

  • object: The input object to be augmented. It is passed to _raw_transform and can be of any type as defined by the augmentation process.

Usage

  • Purpose: Generate augmented variations and select a subset based on a given selection criteria.

Example

Assuming a subclass implements _raw_transform:

aug = YourAugmentationSubclass(selection_size=0.5)
results = aug.transform('sample text')
Note: - If selection_size is less than 1, it represents a fraction of results to select. - If it is an integer, it specifies the exact number to select.