Feature Package Overview#

The eegdash.features namespace re-exports feature extractors, decorators, and dataset utilities from the underlying submodules so callers can import the most common helpers from a single place. To avoid duplicated documentation in the API reference, the classes themselves are documented in their defining modules (see the links below). This page focuses on the high-level orchestration helpers that only live in the package __init__.

High-level discovery helpers#

eegdash.features.get_all_features() → list[tuple[str, Callable]][source]#

Get a list of all available feature functions.

Scans the feature_bank module for functions that have been decorated with a feature_kind.

Returns:

A list of (name, function) tuples for all discovered feature functions.

Return type:

list of tuple

eegdash.features.get_feature_kind(feature: Callable) → eegdash.features.extractors.MultivariateFeature[source]#

Get the ‘kind’ of a feature function.

Identifies whether a feature is univariate, bivariate, or multivariate using decorators.

Parameters:

feature (callable) – The feature function to inspect.

Returns:

An instance of the feature kind.

Return type:

MultivariateFeature

eegdash.features.get_feature_predecessors(feature_or_extractor: Callable | None) → list[source]#

Get the dependency hierarchy for a feature or feature extractor.

This function recursively traverses the parent_extractor_type attribute of a feature or extractor to build a list representing its dependency lineage.

Parameters:

feature_or_extractor (callable) – The feature function or FeatureExtractor instance to inspect.

Returns:

A nested list representing the dependency tree. For a simple linear chain, this will be a flat list from the specific feature up to the base signal input. For multiple dependencies, it contains tuples of sub-dependencies.

Return type:

list

Notes

The traversal stops when it reaches a predecessor of None, which typically represents the raw signal.

Examples

>>> # Example: Linear dependency with a branching dependency
>>> print(get_feature_predecessors(feature_bank.spectral_entropy))
    [<function spectral_entropy at 0x...>,
    <function spectral_normalized_preprocessor at 0x...>,
    <function spectral_preprocessor at 0x...>,
    (None, [<function signal_hilbert_preprocessor at 0x...>, None])]
eegdash.features.get_all_feature_kinds() → list[tuple[str, type[TypeAliasForwardRef('eegdash.features.extractors.MultivariateFeature')]]][source]#

Get a list of all available feature ‘kind’ classes.

Scans the kinds module for all classes that subclass MultivariateFeature.

Returns:

A list of (name, class) tuples for all discovered feature kinds.

Return type:

list of tuple

eegdash.features.get_all_preprocessor_output_types() → list[tuple[str, type[BasePreprocessorOutputType]]][source]#

Get a list of all available preprocessor output type classes.

Scans the feature_bank module for all classes that subclass BasePreprocessorOutputType.

Returns:

A list of (name, class) tuples for all discovered preprocessor output types.

Return type:

list of tuple

Dataset and extraction utilities#

eegdash.features.extract_features(concat_dataset: BaseConcatDataset, features: FeatureExtractor | Dict[str, Callable] | List[Callable], *, batch_size: int = 512, n_jobs: int = 1) → FeaturesConcatDataset[source]#

Extract features from a collection of windowed recordings.

This function applies a feature extraction pipeline to every individual recording in a BaseConcatDataset.

Parameters:

  • concat_dataset (BaseConcatDataset) – A concatenated dataset of WindowsDataset or EEGWindowsDataset instances.

  • features (FeatureExtractor or dict or list) – The feature extractor(s) to apply. Can be a FeatureExtractor instance, a dictionary of named feature functions, or a list of feature functions.

  • batch_size (int, default 512) – The size of batches used for feature extraction within each recording.

  • n_jobs (int, default 1) – The number of parallel jobs to use for processing different recordings simultaneously.

Returns:

A unified collection of feature datasets corresponding to the input recordings.

Return type:

FeaturesConcatDataset

eegdash.features.fit_feature_extractors(concat_dataset: BaseConcatDataset, features: FeatureExtractor | Dict[str, Callable] | List[Callable], batch_size: int = 8192) → FeatureExtractor[source]#

Fit trainable feature extractors on a concatenated dataset.

Scans the provided feature pipeline for components that require training (subclasses of TrainableFeature). If found, the function iterates through the dataset in batches to perform partial fitting before finalization.

Parameters:

  • concat_dataset (BaseConcatDataset) – The dataset used to train the feature extractors.

  • features (FeatureExtractor or dict or list) – The feature extractor pipeline(s) to fit.

  • batch_size (int, default 8192) – The batch size to use when streaming data through the partial_fit() phase.

Returns:

The fitted feature extractor instance, ready for feature extraction.

Return type:

FeatureExtractor

Notes

If the provided extractors are not trainable, the function returns the original input without modification.

eegdash.features.load_features_concat_dataset(path: str | Path, ids_to_load: list[int] | None = None, n_jobs: int = 1) → eegdash.features.datasets.FeaturesConcatDataset[source]#

Load a stored FeaturesConcatDataset from a directory.

This function reconstructs a concatenated dataset by loading individual FeaturesDataset instances from numbered subdirectories.

Parameters:

  • path (str or pathlib.Path) – The root directory where the dataset was previously saved. This directory should contain numbered subdirectories.

  • ids_to_load (list of int, optional) – A list of specific recording IDs (subdirectory names) to load. If None, all numbered subdirectories found in the path are loaded in ascending numerical order.

  • n_jobs (int, default=1) – The number of CPU cores to use for parallel loading. Set to -1 to use all available processors.

Returns:

A unified concatenated dataset containing the loaded recordings.

Return type:

FeaturesConcatDataset

See also

braindecode.datautil.load_concat_dataset

Notes

The function expects the directory structure generated by FeaturesConcatDataset.save(). It automatically reconstructs the feature DataFrames (safetensors), metadata (Pickle), recording info (FIF), and preprocessing keyword arguments (JSON).

See also#

  • eegdash.features.extractors for the feature-extraction base classes such as FeatureExtractor.

  • eegdash.features.datasets for dataset wrappers like FeaturesConcatDataset.

  • eegdash.features.feature_bank.* for the concrete feature families (complexity, connectivity, spectral, and more).