| name | lemma-model |
| description | Final modeling once the baseline and feature set are locked: tune against validation, audit overfitting, touch the test set once, justify the complexity. |
| homepage | https://github.com/tkpratardan/lemma |
| license | MIT |
Final modeling
Deliver
Provide a comparison against the baseline, the selected model and threshold,
stability and subgroup evidence, a locked final evaluation, and operational
limitations.
Check
- Keep one validation design and decision metric across candidates.
- Compare a small motivated model set; keep preprocessing and tuning within
training data and set fixed seeds.
- Diagnose train-validation gaps, fold or temporal stability, calibration,
subgroup behavior, latency, and interpretability.
- Challenge the most fragile feature or assumption with an ablation, shifted
window, or leakage audit.
- Lock the pipeline and threshold before touching the final test set once.
Do not tune against the test set, change metrics mid-search, compare different
splits, omit the baseline, or hide subgroup regressions behind an aggregate
gain.
For tuning and error-analysis detail, read
references/deep-guide.md.
