DriftPolicy#

class scikitplot.feature_selection.DriftPolicy(n_bins=10, max_categories=50, epsilon=0.0001, psi_warn=0.1, psi_alert=0.25, min_count=30)[source]#

How a drift reference is frozen and judged.

Parameters:

n_binsint, default=10

Quantile bins of a numeric reference (ties can merge bins).

max_categoriesint, default=50

Most frequent training categories kept by name; the rest share one "other" bucket.

epsilonfloat, default=1e-4

Floor applied to bucket proportions before the logarithm of the PSI, so an empty bucket gives a large finite value instead of infinity.

psi_warn, psi_alertfloat, default=0.1, 0.25

Population stability index thresholds for the warn and drift statuses (the conventional 0.1 / 0.25 bands).

min_countint, default=30

Fewest current non-missing values to judge a feature; below it the status is insufficient.

Parameters:

• n_bins (int)

• max_categories (int)

• epsilon (float)

• psi_warn (float)

• psi_alert (float)

• min_count (int)

epsilon: float = 0.0001#

max_categories: int = 50#

min_count: int = 30#

n_bins: int = 10#

psi_alert: float = 0.25#

psi_warn: float = 0.1#

to_dict()[source]#

Return the config form.

Return type:

dict[str, Any]