vertebrae.extractors.callable

Custom callable feature extractor.

Classes

CallableExtractor

Wrap a custom Python callable as a feature extractor.

Module Contents

class vertebrae.extractors.callable.CallableExtractor(name, transform_fn, fit_fn=None, modality='unknown', extractor_type='custom_callable', recipe_data=None, allow_sparse=True, streaming_safe=False, resource_profile_adapter=None, cache_identity=None)[source]

Wrap a custom Python callable as a feature extractor.

Parameters:
  • name (str) – User-facing extractor name.

  • transform_fn (Callable[[Any], Any]) – Callable that converts inputs into dense or sparse embeddings.

  • fit_fn (Optional[Callable[[Any, Any], Any]]) – Optional callable invoked during fit.

  • modality (str) – Input modality metadata.

  • extractor_type (str) – Extractor family metadata.

  • recipe_data (Optional[Dict[str, Any]]) – Extra serializable data to include in recipe().

  • allow_sparse (bool) – Whether sparse transform outputs are allowed.

  • streaming_safe (bool) – Whether independent batches can be transformed without access to the full dataset.

  • resource_profile_adapter (Optional[Any])

  • cache_identity (Optional[str])

fit(X, y=None)[source]

Fit the callable extractor when a fit function is supplied.

Parameters:
  • X (Any) – Input samples.

  • y (Any) – Optional labels.

Returns:

This extractor.

Return type:

CallableExtractor

transform(X)[source]

Transform inputs with transform_fn and validate embeddings.

Parameters:

X (Any) – Input samples.

Returns:

Dense or sparse numeric embedding matrix.

Return type:

numpy.ndarray

fit_transform(X, y=None)[source]

Fit the extractor and transform inputs.

Parameters:
  • X (Any) – Input samples.

  • y (Any) – Optional labels.

Returns:

Dense or sparse numeric embedding matrix.

Return type:

numpy.ndarray

recipe()[source]

Return a serializable callable extractor recipe.

Returns:

JSON-compatible recipe dictionary.

Return type:

Dict[str, Any]

get_resource_profile_adapter()[source]

Return the optional user-supplied resource profiling adapter.

Return type:

Any