cardinal.base

Classes

class cardinal.base.BaseQuerySampler(batch_size: int)

Abstract Base Class for query samplers

A query sampler is an object that takes as input labeled and/or unlabeled samples and use knowledge from them to selected the most informative ones.

Parameters

batch_size – Numbers of samples to select.

abstract fit(X: ndarray, y: Optional[ndarray] = None)

Fit the model on labeled samples.

Parameters

• X – Labeled samples of shape (n_samples, n_features).

• y – Labels of shape (n_samples).

Returns

The object itself

abstract select_samples(X: array) → array

Selects the samples to annotate from unlabeled data.

Parameters

X – Pool of unlabeled samples of shape (n_samples, n_features).

Returns

Indices of the selected samples of shape (batch_size).

class cardinal.base.ScoredQuerySampler(batch_size: int, strategy: str = 'top', random_state: Optional[Union[RandomState, int]] = None)

Abstract Base Class handling query samplers relying on a total order. Query sampling methods often scores all the samples and then pick samples using these scores. This base class handles the selection system, only a scoring method is then required.

Parameters

• batch_size – Numbers of samples to select.

• strategy – Describes how to select the samples based on scores. Can be “top”, “weighted”.

• random_state – Random seeding

abstract score_samples(X: array) → array

Give an informativeness score to unlabeled samples.

Parameters

X – Samples to evaluate.

Returns

Scores of the samples.

select_samples(X: array) → array

Selects the samples from unlabeled data using the internal scoring.

Parameters

• X – Pool of unlabeled samples of shape (n_samples, n_features).

• strategy – Strategy to use to select queries. Can be one of top, linear_choice, or squared_choice.

Returns

Indices of the selected samples of shape (batch_size).