tarexp.component.sampler module

TARexp provides reference implementations of a variety of TAR-specific algorithms, to aid reproducibility and reduce experimenter work. For instance, uncertainty sampling [1], relevance feedback [2], and simple random sampling batch selection algorithms are provided.

This interface is inspired by libact, an Active Learning library.

See also

[1]

David D. Lewis, and William A. Gale. “A sequential algorithm for training text classifiers.” SIGIR 1994. Springer, London, 1994. https://arxiv.org/abs/cmp-lg/9407020

[2]

Rocchio Algorithm for psuedo relevance feedback https://en.wikipedia.org/wiki/Rocchio_algorithm

class tarexp.component.sampler.Sampler(**kwargs)

Bases: Component

sampleDocs(nask: int, ledger: Ledger, scores, *args, **kwargs)

class tarexp.component.sampler.UncertaintySampler(**kwargs)

Bases: Sampler

Implementing least confidence uncertainty sampling.

sampleDocs(nask: int, ledger: Ledger, scores, **kwargs)

freeze()

class tarexp.component.sampler.RelevanceSampler(**kwargs)

Bases: Sampler

sampleDocs(nask: int, ledger: Ledger, scores, **kwargs)

class tarexp.component.sampler.RandomSampler(**kwargs)

Bases: Sampler

begin(random_seed=None, *args, **kwargs)

sampleDocs(nask: int, ledger: Ledger, scores=None, dist=None, **kwargs)