incerto.active.CoreSetSelection

incerto.active.CoreSetSelection#

class incerto.active.CoreSetSelection(batch_size=100)[source]#

Bases: object

Core-Set selection for active learning.

Selects samples that best represent the overall data distribution using k-center greedy algorithm.

Reference:

Sener & Savarese, “Active Learning for Convolutional Neural Networks: A Core-Set Approach” (ICLR 2018)

Parameters:

batch_size (int) – Number of samples to select

__init__(batch_size=100)[source]#
Parameters:

batch_size (int)

Methods

__init__([batch_size])

query(model, x_unlabeled[, x_labeled, ...])

Select core-set using greedy k-center.
__init__(batch_size=100)[source]#
Parameters:

batch_size (int)

query(model, x_unlabeled, x_labeled=None, features_unlabeled=None, features_labeled=None)[source]#

Select core-set using greedy k-center.

Parameters:

  • model (Module) – Model for feature extraction

  • x_unlabeled (Tensor) – Unlabeled data

  • x_labeled (Optional[Tensor]) – Labeled data (optional)

  • features_unlabeled (Optional[Tensor]) – Precomputed features for unlabeled data

  • features_labeled (Optional[Tensor]) – Precomputed features for labeled data

Return type:

Tensor

Returns:

Indices of selected samples

Contents