feature-discovery

name	feature-discovery
description	Build candidate features, run feature screening and diagnostics, and prepare accepted feature sets for downstream flywheel experiments.

Feature Discovery

Use this skill when iterating on feature engineering and deciding which candidates are safe/useful for experiments.

What this skill does

1. Builds/loads feature candidate matrices.
2. Applies quality and collinearity screening.
3. Runs importance diagnostics (MDI/MDA/SFI) where needed.
4. Produces a selected feature list and rejection rationale.

Core Python APIs

• openquant.feature_diagnostics.feature_screen_report(...)
• openquant.feature_diagnostics.mdi_importance(...)
• openquant.feature_diagnostics.mda_importance(...)
• openquant.feature_diagnostics.sfi_importance(...)
• openquant.feature_diagnostics.substitution_effect_report(...)

Minimal workflow snippet

import openquant
import polars as pl

screen = openquant.feature_diagnostics.feature_screen_report(
    X=feature_df,
    min_coverage=0.95,
    max_corr=0.92,
)
selected = screen["selected_features"]
reasons = screen["rejection_reasons"]

Notebook target

• notebooks/python/07_feature_engineering_discovery_loop.ipynb

Validation

uv run --python .venv/bin/python pytest python/tests/test_feature_diagnostics_module.py python/tests/test_research_grid_and_screening.py -q