Notes for parameter ranges in Trees
(param: type, range)
learning_rate: float, (0.0, inf)
n_estimators: int, [1, inf)
min_samples_split: int -> [2, inf], float-> (0, 1]
min_samples_leaf: int -> [1, inf], float-> (0, 1]
min_weight_fraction_leaf: float-> [0, 0.5]
max_depth: int, if not none, then [1, inf)
min_impurity_decrease: float, [0, inf)
subsample: float, (0, 1]
alpha: float, (0, 1)
max_features: int -> [1, number of features], float -> (0, 1], or string (not checked with check_scalar)
ccp_alpha: float, [0, inf)
verbose: int, [0, inf), or np.bool_
max_leaf_nodes: int, if not none, then [2, inf)
warm_start: check that it is int, (inf, inf) or np.bool_
validation_fraction: float, (0, 1)
n_iter_no_change: int -> [1, inf)
tol: float, (0, inf)

Hi @glemaitre: just want to note that for the 7 remaining params in the PR task list (and also max_features), the validation is coming from BaseDecisionTree.



I have added tests for these params in BaseGradientBoosting that fail right now, but they should pass once BaseDecisionTree #21990 gets merged in. (they are commented out for now)

For these remaining params, please let me know if it would be helpful to include the validation explicitly within BaseGradientBoosting, too. Thank you!

Hi @glemaitre and @ogrisel, please review this at you convenience. Thank you!