| name | model-training |
| description | Model training for data scientists — delegates to ds-model-trainer agent. Use
when training classification or regression models, tuning hyperparameters with
Optuna, running cross-validation, or comparing multiple algorithms.
|
Model Training Command
Train, tune, and compare ML models with cross-validation and experiment logging
Usage
/ds-model-training <description-or-file>
Examples
/ds-model-training "Binary classification for customer churn with XGBoost"
/ds-model-training "Regression model for house price prediction — tune with Optuna"
/ds-model-training "Compare RandomForest vs LightGBM vs LogisticRegression on this dataset"
/ds-model-training notebooks/training_spec.md
What This Command Does
- Invokes the ds-model-trainer agent
- Reads dataset description or spec to identify problem type and target
- Loads KB patterns from
scikit-learn and xgboost domains
- Generates:
- Baseline model (DummyClassifier / mean predictor)
- Primary model with cross-validation (stratified KFold or TimeSeriesSplit)
- Optuna hyperparameter search with MLflow nested run logging
- Model comparison table (AUC, F1, RMSE per fold)
Agent Delegation
| Agent | Role |
|---|
ds-model-trainer | Primary — estimators, cross-val, Optuna tuning, model comparison |
ds-experiment-tracker | Escalation — when runs must be logged to MLflow |
ds-model-evaluator | Escalation — when full evaluation suite is needed after training |
ds-time-series-analyst | Escalation — when target is a time series |
KB Domains Used
scikit-learn — estimators, cross-validation, model selection
xgboost — XGBoost and LightGBM training patterns
mlflow — autologging, run logging, experiment tracking
Output
The agent generates training code with cross-validation, a model comparison table, and the best-performing pipeline object ready for evaluation.