glm-factor-optimizer Documentation¶

glm-factor-optimizer helps you design GLM factors in a way that is easy to inspect, rerun, and save. Use it to learn numeric bins, group categorical levels, rank candidate factors, compare validation results, and preserve the choices as reviewable artifacts.

For LLM-assisted workflows, start with the LLM Quickstart, the sitemap, or the plain-text llms.txt project map.

A typical modeling session looks like this:

1. Split data into train, validation, and holdout.
2. Rank coarse candidate factors.
3. Work factor by factor with FactorBlock.
4. Accept bins/groups deliberately after inspection.
5. Fit the current main-effects model.
6. Refine accepted factors with the full model fixed.
7. Search and test candidate interactions.
8. Finalize once on holdout.
9. Save specs, history, metrics, and validation tables.

Documentation Map¶

Tutorial:

• LLM Quickstart
• Notebook GLM Factor Study Workflow

Discovery:

• Sitemap
• llms.txt
• llms-full.txt

How-to guides:

• Rank and Screen Candidate Factors
• Refine Factors With the Full Model Fixed
• Test and Accept Interactions
• Run an Automatic Baseline Workflow
• Save and Audit a Study

Reference:

• Public API Reference
• Binning and Grouping Specs
• Validation Outputs

Explanation:

• Architecture
• Modeling Principles

Main Interfaces¶

Use GLMStudy for iterative notebook modeling:

from glm_factor_optimizer import GLMStudy

study = GLMStudy(
    df,
    target="events",
    exposure="hours",
    prediction="predicted_count",
)

Use RateGLM or GLM for low-level manual work:

from glm_factor_optimizer import RateGLM

glm = RateGLM(target="events", exposure="hours")

Use GLMWorkflow for compact batch baselines:

from glm_factor_optimizer import GLMWorkflow

Use glm_factor_optimizer.spark for Spark-oriented primitives.