incerto.active.UncertaintySampling

incerto.active.UncertaintySampling#

class incerto.active.UncertaintySampling(acquisition_fn, batch_size=100)[source]#

Bases: object

Uncertainty-based sampling strategy.

Selects samples with highest acquisition scores.

Parameters:

  • acquisition_fn (BaseAcquisition) – Acquisition function to use

  • batch_size (int) – Number of samples to select per query

Example

>>> acquisition = BALDAcquisition(num_samples=20)
>>> strategy = UncertaintySampling(acquisition, batch_size=100)
>>> indices = strategy.query(model, unlabeled_data)
__init__(acquisition_fn, batch_size=100)[source]#
Parameters:

  • acquisition_fn (BaseAcquisition)

  • batch_size (int)

Methods

__init__(acquisition_fn[, batch_size])

query(model, x_unlabeled)

Query samples based on uncertainty.
__init__(acquisition_fn, batch_size=100)[source]#
Parameters:

  • acquisition_fn (BaseAcquisition)

  • batch_size (int)

query(model, x_unlabeled)[source]#

Query samples based on uncertainty.

Parameters:

  • model (Module) – Trained model

  • x_unlabeled (Tensor) – Unlabeled data (N, *)

Return type:

Tensor

Returns:

Indices of selected samples (batch_size,)

Contents