autoxgb: autoxgb strategy

In Prometheus77/ezmlr: Easy Interface to mlr

Description

The autoxgb strategy creates an XGBoost learner to fit a specific task. It performs imputation if any variables are missing, and one-hot encoding if any factor variables are present. By default, it uses the following hyperparameters:

• nrounds = 200

• eta = 2^-6

• max_depth = 3

• colsample_bytree = 0.6

• subsample = 0.6

Available parameters for use in in strategy.pars:

• par.vals (list) List of hyperparameter values to override defaults.

• impute.pars (list) List of parameters to pass to makeImputeWrapper

• dummy.pars (list) List of parameters to pass to makeDummyFeaturesWrapper

The autoxgb_tuned strategy creates an XGBoost learner to fit a specific task and tunes it for improved performance.

Available parameters for use in in strategy.pars:

• impute.pars (list) List of parameters to pass to makeImputeWrapper

• dummy.pars (list) List of parameters to pass to makeDummyFeaturesWrapper

• tune.pars (list) list of parameters to pass to makeTuneWrapper

Prometheus77/ezmlr documentation built on May 17, 2019, 6:34 p.m.