Examples · Bands

governance_bands

The Welch t-test variant: adjacent bands must be statistically distinguishable before they are allowed to exist.

Code#

governance_bands.py

from _setup import latent, y_train
 
from compileml.bands import governance_bands, semantic_bands
 
committee = governance_bands(latent, y_train, max_bands=10, min_band_size=2000)
discovered = semantic_bands(latent, y_train, max_bands=10, min_band_size=300)
 
print("governance_bands :", len(committee.labels), "bands —", committee.labels)
print("semantic_bands   :", len(discovered.labels), "bands —", discovered.labels)
print()
print("Larger minimum band size and a stricter separation threshold buy")
print("stability across refits, at the cost of resolution.")

Output#

Captured from an actual run against compileml 0.9.0 and the UCI credit panel. If this script stops working, the build fails.

captured in CIgovernance_bands.py

governance_bands : 9 bands — ['G01', 'G02', 'G03', 'G04', 'G05', 'G06', 'G07', 'G08', 'G09']
semantic_bands   : 1 bands — ['G01']
 
Larger minimum band size and a stricter separation threshold buy
stability across refits, at the cost of resolution.

Notes#

Tighter alpha means fewer, wider, better-evidenced bands.

API reference: governance_bands →

Related notebook