Extra Learners For mlr3 • mlr3extralearners

Extra Learners for mlr3

What is mlr3extralearners?

This mlr3 extension contains all mlr3 learners that are not in mlr3learners or the core packages. Besides, it contains helper functions to list all learners from the mlr3verse and install their required packages. See the interactive learner list for the full list of learners in the mlr3verse.

list_mlr3learners(select = c("id", "mlr3_package", "required_packages"))
#>                      id      mlr3_package      required_packages
#>   1: classif.AdaBoostM1 mlr3extralearners                  RWeka
#>   2:       classif.bart mlr3extralearners                 dbarts
#>   3:        classif.C50 mlr3extralearners                    C50
#>   4:   classif.catboost mlr3extralearners               catboost
#>   5:    classif.cforest mlr3extralearners partykit,sandwich,coin
#>  ---
#> 128:        surv.ranger      mlr3learners                 ranger
#> 129:         surv.rfsrc mlr3extralearners randomForestSRC,pracma
#> 130:         surv.rpart         mlr3proba  rpart,distr6,survival
#> 131:           surv.svm mlr3extralearners            survivalsvm
#> 132:       surv.xgboost      mlr3learners                xgboost

mlr3extralearners lives on GitHub and will not be on CRAN. This enables us to include packages such as catboost, which is not on CRAN either.

Install with:

remotes::install_github("mlr-org/mlr3extralearners")

Installing and Loading Learners

The package includes functionality for detecting if you have the required packages installed to use a learner, and ships with the function install_learner which can install all required learner dependencies.

lrn("regr.gbm")
#> Package 'gbm' required but not installed for Learner 'regr.gbm'

install_learners("regr.gbm")
lrn("regr.gbm")
#> <LearnerRegrGBM:regr.gbm>
#> * Model: -
#> * Parameters: keep.data=FALSE, n.cores=1
#> * Packages: gbm
#> * Predict Type: response
#> * Feature types: integer, numeric, factor, ordered
#> * Properties: importance, missings, weights

Extending mlr3extralearners

New learners can be created with the create_learner function. This assumes you have a local copy of mlr3extralearners. This function will automatically create the learner, learner tests, parameter tests and update the DESCRIPTION if required. Once all tests are passing locally, open a pull request with the “New Learner” template. More detailed instructions can be found in the mlr3 book.

library(mlr3proba)
create_learner(
  classname = "Locfit",
  algorithm = "localised fit",
  type = "dens",
  key = "locfit",
  package = "locfit",
  caller = "density.lf",
  feature_types = c("integer", "numeric"),
  predict_types = c("pdf", "cdf"),
  properties = NULL,
  references = FALSE,
  gh_name = "RaphaelS1"
)