Skip to contents

Boosting uses gbm3::gbmt() for binary, count, continuous, and survival outcomes. Existing public argument names are translated internally.

Details

Additional arguments passed through ... can include shrinkage (default 0.1), n.minobsinnode (10), bag.fraction (0.5), train.fraction (1), var.monotone, mFeatures, keep.data, class.stratify.cv, and fold.id. Set n.cores to select the number of GBM3 threads, or supply par.details = gbm3::gbmParallel(...); do not supply both. With neither argument, the gbm.parallel option is respected and the fallback is one thread.

GBM3's learning-rate and training-sample defaults are overridden to retain precmed's previous GBM settings. Fits use the GBMFit class rather than gbm; use the generic predict() method for prediction. Predictions exclude exposure offsets, preserving the previous rate-scale behavior. Results may differ from GBM even with the same settings and seed.