domain-tabular
Skill8 starsChanged 4 months agoMITView on GitHub
Use for structured/tabular data prediction: churn, fraud, credit/risk scoring, price/demand regression, conversion, recommendation features, any CSV/database ML. Picks gradient boosting vs deep tabular, robust feature engineering, correct CV, calibration, and imbalance handling. Triggers on 'predict churn/fraud/price/risk', 'classify this CSV/table', 'tabular', 'XGBoost/LightGBM', 'feature engineering'.
In this repository22 other files it gives its agents.
- accuracy-improvement-loopSkill
- data-rigor-and-leakageSkill
- dataset-profilingSkill
- deployment-explainabilitySkill
- domain-3dSkill
- domain-anomaly-detectionSkill
- domain-audio-speechSkill
- domain-computer-visionSkill
- domain-generativeSkill
- domain-graphSkill
- domain-medical-imagingSkill
- domain-multimodalSkill
- domain-nlp-llmSkill
- domain-recommenderSkill
- domain-reinforcement-learningSkill
- domain-time-seriesSkill
- domain-videoSkill
- literature-reviewSkill
- ml-research-methodologySkill
- notebook-deliverySkill
- rigorous-evaluationSkill
- training-optimizationSkill
The file
Files
Tabular ML - Method Selection
For most tabular problems, gradient-boosted trees beat deep learning - start there.
Decision table
| Situation | Recommended | Notes |
| Default (any size, mixed types) | LightGBM / XGBoost / CatBoost | CatBoost handles categoricals natively; LightGBM is fast; XGBoost robust. Tune with Optuna. |
| Strong baseline / linear signal | Regularized Linear/Logistic, ElasticNet | Fast sanity baseline; interpretable. |
| Very large + rich, or embeddings needed | Deep tabular (FT-Transformer / TabNet) | Only when GBTs plateau and data is large. |
| Tiny data | Linear / simple trees + heavy regularization | Avoid overfitting; strong CV. |
| Ensembling | Blend/stack GBTs + linear | Usually the last real gain. |
Feature engineering & correctness
- Target leakage is the #1 killer: drop features known only after the outcome or that encode the label. Audit each feature's availability at prediction time.
- Fit encoders/scalers/imputers/target-encoding inside CV folds (on train only) - fitting on the full set leaks (
data-rigor-and-leakage). - Handle categoricals (native/CatBoost, or target/one-hot with fold-safe encoding), missing values, and skew deliberately.
- Time-aware data needs a time-based split (train earlier than test), never random.
Evaluation & imbalance
- CV: StratifiedKFold (classification), GroupKFold if entities repeat, TimeSeriesSplit if temporal.
- Imbalanced (fraud/churn): PR-AUC, recall at a fixed precision, class weights or focal/
scale_pos_weight- not raw accuracy. - Calibrate probabilities (Platt/isotonic) if scores drive decisions/thresholds.
- Explainability: SHAP for global + per-prediction attribution (great for stakeholder trust).
- Improve: use
accuracy-improvement-loop(feature engineering usually beats model swaps; then tuning, then stacking).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.