How To Refine Factors With the Full Model Fixed¶

After you have a main-effects model, re-check important factors inside that model. GLMStudy.refine_factor() re-optimizes one accepted factor while keeping all other accepted factors fixed.

Accept an Initial Factor¶

age = study.factor("machine_age", kind="numeric")
age.optimize(trials=100, max_bins=6, n_prebins=12)
age.accept(comment="Initial machine_age factor")

Add more factors:

equipment = study.factor("equipment_type", kind="categorical")
equipment.optimize(trials=100)
equipment.accept(comment="Equipment type grouping")

study.fit_main_effects()

Refine One Factor¶

refined_age = study.refine_factor(
    "machine_age",
    trials=200,
    max_bins=6,
    n_prebins=16,
)

refined_age.compare()
refined_age.bin_table()
refined_age.validation_table()

The refinement keeps every other accepted factor fixed. If the current accepted factor is already in the model, the old version is removed from the fixed set while the proposed replacement is evaluated.

Accept or Leave as Proposal¶

refined_age.accept(comment="Accepted full-model machine_age refinement")

If the proposal is not better or not stable:

refined_age.reject(comment="Rejected because validation gain was too small")

Refine All Factors¶

proposals = study.refine_all(trials=50, accept=False)

This returns one FactorBlock per accepted raw factor. Review each block before accepting. Automatic acceptance is available:

study.refine_all(trials=50, accept=True)

Use automatic acceptance only for exploratory baselines or controlled batch experiments.