BinningPolicy#

class scikitplot.feature_selection.BinningPolicy(enabled=True, rule='auto', n_bins=None, min_bins=2, max_bins=64, min_count=30, strategy='uniform')[source]#

How bin recommendations are computed for numeric features.

Parameters:

enabledbool, default=True

Compute a BinningRecommendation for each feature while profiling.

rule{“auto”, “knuth”, “fd”, “scott”, “sturges”, “doane”, “rice”, “sqrt”, “stone”, “fixed”}, default=”auto”

Bin-count rule. "knuth" maximises Knuth’s Bayesian posterior over every integer count in [min_bins, max_bins] (an exhaustive, exact search); "fixed" uses n_bins; the others are the rules of numpy.histogram_bin_edges.

n_binsint or None, default=None

Bin count for rule="fixed"; must be None for other rules.

min_bins, max_binsint, default=2, 64

Bounds applied to every rule’s count (a clipped count is reported).

min_countint, default=30

Fewest finite values for a recommendation; below it the suitability is unknown.

strategy{“uniform”, “quantile”}, default=”uniform”

Edge placement. The data-driven rules assume equal widths; with "quantile" their count is used with quantile edges, and tied quantiles can reduce the count (reported).

Parameters:

• enabled (bool)

• rule (str)

• n_bins (int | None)

• min_bins (int)

• max_bins (int)

• min_count (int)

• strategy (str)

Notes

User note. A recommendation is advice: it never changes your data. Turn it into a transformer with binning_plan.

enabled: bool = True#

max_bins: int = 64#

min_bins: int = 2#

min_count: int = 30#

n_bins: int | None = None#

rule: str = 'auto'#

strategy: str = 'uniform'#

to_dict()[source]#

Return the config form.

Return type:

dict[str, Any]