Factor Ranking SAGE

Use this skill to rank local quantitative factors with one supported method per run.

Core Workflow

1. Read references/input_schema.md before preparing the input JSON.
2. Provide local factor and label CSV files with one unique row per (date, ticker) observation.
3. Configure one fixed train/validation split and an embargo suitable for the label horizon.
4. Choose one mode:
   • Use mode=mrmr for deterministic regression F-statistic relevance and Pearson-redundancy Top-K selection on training rows.
   • Use mode=sage for fixed-model Marginal-SAGE MSE contribution. Choose model.type=lgbm or model.type=mlp.
5. Run the matching smoke configuration first:

python scripts/run_factor_selection.py --input examples/factor_selection_mrmr_smoke.json
python scripts/run_factor_selection.py --input examples/factor_selection_lgbm_smoke.json
python scripts/run_factor_selection.py --input examples/factor_selection_mlp_smoke.json