tune

Hyperparameter search over a parameter grid. Cross-validated internally. Returns the best model and the full trial history.

Signature

ml.tune(data, target, *, algorithm=None, params=None, method="random", n_trials=20, cv_folds=3, seed)

ml_tune(data, target, algorithm = NULL, n_trials = 20L, cv_folds = 3L, method = "random", seed = NULL, params = NULL)

Parameters

Parameter	Type	Default	Description
`data`	DataFrame	—	Training data
`target`	str	—	Target column
`algorithm`	str \| None	`None`	Algorithm to tune. `None` uses the model's algorithm or `"auto"`.
`params`	dict \| None	`None`	Parameter ranges. `None` uses sensible defaults.
`method`	str	`"random"`	`"random"` or `"grid"`
`n_trials`	int	`20`	Number of search trials
`cv_folds`	int	`3`	Cross-validation folds per trial
`seed`	int	—	Random seed

Returns

TuningResult with:

.best_model — the model with the best CV score
.best_params_ — winning hyperparameters
.trials — DataFrame of all trials and scores

Examples

Default tuning

result = ml.tune(s.train, "target", algorithm="xgboost", seed=42)
print(result.best_params_)
ml.evaluate(result.best_model, s.valid)

result <- ml_tune(s$train, "target", algorithm = "xgboost", seed = 42)
result$best_params_
ml_evaluate(result$best_model, s$valid)

Custom parameter grid

result = ml.tune(
    s.train, "target",
    algorithm="random_forest",
    params={
        "n_estimators": [100, 200, 500],
        "max_depth": [5, 10, 20, None],
    },
    method="grid",
    seed=42,
)

result <- ml_tune(
  s$train, "target",
  algorithm = "random_forest",
  params = list(
    n_estimators = c(100, 200, 500),
    max_depth = c(5, 10, 20)
  ),
  method = "grid",
  seed = 42
)

stack →