eegdash.features.trainable#

Core Trainable Feature Interface.

This module defines the interface for creating trainable features.

The module provides the base class:

  • TrainableFeature - The interface for features requiring a fitting phase.

Classes

TrainableFeature()

Abstract base class for features requiring a training phase.
class eegdash.features.trainable.TrainableFeature[source]

Bases: ABC

Abstract base class for features requiring a training phase.

This class provides the interface for features that must be fitted on a representative dataset before they can process new samples.

_is_trained

Internal flag indicating whether the feature has completed its training phase.

Type:

bool

abstractmethod clear()[source]

Reset the internal state of the feature.

This method must be implemented by subclasses to clear any learned parameters, statistics, or buffers.

abstractmethod partial_fit(*x, y=None)[source]

Update the extractor’s state using a single batch of data.

This method allows for incremental learning, making it possible to train on datasets that are too large to fit into memory at once.

Parameters:

  • *x (tuple of ndarray) – The input data batch.

  • y (ndarray, optional) – Target labels associated with the batch, required for supervised feature extraction methods.

fit()[source]

Finalize the training of the feature extractor.

This method should be called after the entire training set has been processed via partial_fit(). It transitions the object to a “trained” state, enabling the __call__() method.