band_efficiency

def band_efficiency(latent, y, bands, *, scale: int = 1000, n_boot: int = 200, alpha: float = 0.05, seed: int = 7) -> dict

from compileml.bands import band_efficiency

Quantify the discrimination cost of a band ladder.

Args: latent: Continuous latents in [0, 1] (whitebox predictions, clipped). y: Binary outcomes aligned with latent. bands: A compiled artifact dict, a :class:`BandSpec`, or a sequence of float band edges. Assignment always runs on the fixed-point integer ladder — the deployed semantics. scale: Display scale used when bands is not an artifact. n_boot: Bootstrap resamples for the per-band AUC intervals.

Returns: dict with continuous_gini, band_ordinal_gini, gini_gap (the money on the table, in Gini points), gini_gap_pct, a per_band table (n, bad_rate, within-band AUC + CI), and worst_band — the band whose lower CI bound sits highest above 0.5, i.e. the strongest refinement candidate.

Parameters#

Name	Type	Default	Kind
latent	—	required	positional
y	—	required	positional
bands	—	required	positional
scale	int	1000	keyword-only
n_boot	int	200	keyword-only
alpha	float	0.05	keyword-only
seed	int	7	keyword-only

Returns#

dict

Raises#

ValueError

Read from the function body and the private helpers it calls, not inferred. A function that raises nothing has no section here.

Example#

python

from compileml.bands import band_efficiency
 
report = band_efficiency(latent, y_train, bands)
print(report["verdict"])  # refinable | exhausted | inconclusive

Worked example: band_efficiency →

View source