anndict.utils.stable_label

anndict.utils.stable_label(x, y, classifier, *, max_iterations=100, stability_threshold=0.05, moving_average_length=3, random_state=None)

Trains a classifier using a semi-supervised approach where labels are probabilistically reassigned based on classifier predictions.

Parameters:

x ndarray

feature matrix.

y ndarray

initial labels for all data.

classifier ClassifierMixin

a classifier instance that implements fit and predict_proba methods.

max_iterations int (default: 100)

maximum number of iterations for updating labels.

stability_threshold float (default: 0.05)

threshold for the fraction of labels changing to consider the labeling stable.

moving_average_length int (default: 3)

number of past iterations to consider for moving average.

random_state int | None (default: None)

seed for random number generator for reproducibility.

Return type:

tuple[ClassifierMixin, list[float], int, ndarray]

Returns:

classifierClassifierMixin

trained classifier.

historylist[float]

percentage of labels that changed at each iteration.

iterationsint

number of iterations run.

final_labelsndarray

the labels after the last iteration.

@todo - switch pca_density_subsets to use pca_density_filter_main or pca_density_filter_adata