GBM¶

Supported HTTP methods and descriptions¶

URL¶

http://<h2oHost>:<h2oApiPort>/GBM.json

Input parameters¶

classification, a boolean, expert

Do classification or regression. Since version 1
validation, a Frame, expert

Validation frame. Since version 1
n_folds, a int, expert

Number of folds for cross-validation (if no validation data is specified). Since version 1
holdout_fraction, a float, expert

Fraction of training data (from end) to hold out for validation (if no validation data is specified). Since version 1
keep_cross_validation_splits, a boolean, expert

Keep cross-validation dataset splits. Since version 1
ntrees, a int, critical

Number of trees. Grid Search, comma sep values:50,100,150,200. Since version 1
max_depth, a int, critical

Maximum tree depth. Grid Search, comma sep values:5,7. Since version 1
min_rows, a int, secondary

Fewest allowed observations in a leaf (in R called ‘nodesize’). Grid Search, comma sep values. Since version 1
nbins, a int, secondary

Build a histogram of this many bins, then split at the best point. Since version 1
score_each_iteration, a boolean, expert

Perform scoring after each iteration (can be slow). Since version 1
importance, a boolean, expert

Compute variable importance (true/false).. Since version 1
balance_classes, a boolean, expert

Balance training data class counts via over/under-sampling (for imbalanced data). Since version 1
class_sampling_factors, a float[], secondary

Desired over/under-sampling ratios per class (lexicographic order).. Since version 1
max_after_balance_size, a float, expert

Maximum relative size of the training data after balancing class counts (can be less than 1.0). Since version 1
checkpoint, a Key, expert

Model checkpoint to start building a new model from. Since version 1
overwrite_checkpoint, a boolean, expert

Overwrite checkpoint. Since version 1
family, a Family, critical

Distribution for computing loss function. AUTO selects gaussian for continuous and multinomial for categorical response. Since version 1
learn_rate, a double, secondary

Learning rate, from 0. to 1.0. Since version 1
grid_parallelism, a int, secondary

Grid search parallelism. Since version 1
seed, a long, expert

Seed for the random number generator - only for balancing classes (autogenerated). Since version 1
group_split, a boolean, expert

Perform Group Splitting Categoricals. Since version 1

Output JSON elements¶

xval_models, a Key[]

Cross-validation models. Since version 1, expert
_distribution, a long[]

Class distribution. Since version 1, expert

HTTP response codes¶

200 OK Success and error responses are identical.