make_pairwise_ranker#

scikitplot.corpus.make_pairwise_ranker(score_func, *, greater_is_better=True, use_normalized_text=True, score_cutoff=None, name=None, **kwargs)[source]#

Make a CustomRetrievalIndex ranker from a pairwise function.

Parameters:

score_funccallable: score_func(query, text, **kwargs) -> float, for example edit_similarity, a RapidFuzz ratio or a cross-encoder.
greater_is_betterbool, default=True: False for a distance, which is negated so higher is better.
use_normalized_textbool, default=True: Compare with normalized_text when a document has one.
score_cutofffloat or None, default=None: Minimum (signed) score of a returned hit.
namestr or None, default=None: Recorded as each hit’s native_metric.
**kwargsdict: Passed to score_func on every call.

Returns:

PairwiseRanker

Raises:

TypeError: For a non-callable score_func, non-bool flags, or kwargs that do not fit its signature.
ValueError: For a non-finite score_cutoff.

Parameters:

score_func (Callable[[...], Any])
greater_is_better (bool)
use_normalized_text (bool)
score_cutoff (float | None)
name (str | None)
kwargs (Any)

Return type:

PairwiseRanker

See also

make_levenshtein_ranker: The Levenshtein ranker with backend provenance.

Examples

>>> from scikitplot.corpus import (
...     CorpusDocument,
...     CustomRetrievalIndex,
...     edit_similarity,
...     make_pairwise_ranker,
... )
>>> index = CustomRetrievalIndex(
...     custom_scorer_fn=make_pairwise_ranker(edit_similarity)
... )
>>> index.build(
...     [
...         CorpusDocument.create("a.txt", 0, "kitten"),
...         CorpusDocument.create("b.txt", 0, "sitting"),
...     ]
... )
>>> [hit.doc.text for hit in index.search("kitten")]
['kitten', 'sitting']