Hyperparameter Tuning
The MATIH ML Service provides automated hyperparameter optimization using Ray Tune, supporting Bayesian optimization, grid search, random search, and population-based training strategies.
Tuning API
POST /api/v1/tuning/jobs
Content-Type: application/json
{
"model_class": "sklearn.ensemble.GradientBoostingClassifier",
"search_space": {
"learning_rate": {"type": "loguniform", "min": 0.001, "max": 0.1},
"n_estimators": {"type": "choice", "values": [100, 200, 500]},
"max_depth": {"type": "randint", "min": 3, "max": 12}
},
"search_algorithm": "bayesian",
"num_trials": 50,
"metric": "val_accuracy",
"mode": "max",
"max_concurrent_trials": 4
}Search Algorithms
| Algorithm | Description | Best For |
|---|---|---|
bayesian | Bayesian optimization (Optuna/HyperOpt) | Small-medium search spaces |
grid | Exhaustive grid search | Discrete, small spaces |
random | Random sampling | Large search spaces |
pbt | Population-based training | Neural network schedules |
Source Files
| File | Path |
|---|---|
| HyperparameterTuner | data-plane/ml-service/src/training/hyperparameter_tuner.py |
| Tuning API | data-plane/ml-service/src/api/tuning.py |