mlr3learners.lightgbm

mlr3learners.lightgbm brings the LightGBM gradient booster to the mlr3 framework by using the official lightgbm R implementation.

Features

integrated learner-native cross-validation (CV) using lgb.cv before the actual model training to find the optimal num_iterations for the given training data and parameter set
GPU support

Installation

As of lightgbm version 3.0.0, you can install the mlr3learners.lightgbm R package with:

install.packages("remotes")
# stable version:
remotes::install_github("mlr3learners/mlr3learners.lightgbm")
# dev version:
# remotes::install_github("mlr3learners/mlr3learners.lightgbm@development")

Example

library(mlr3)
library(mlr3learners.lightgbm)
task = mlr3::tsk("iris")
learner = mlr3::lrn("classif.lightgbm", objective = "multiclass")

learner$param_set$values = mlr3misc::insert_named(
  learner$param_set$values,
    list(
    "early_stopping_round" = 10,
    "learning_rate" = 0.1,
    "seed" = 17L,
    "metric" = "multi_logloss",
    "num_iterations" = 100,
    "num_class" = 3
  )
)

learner$train(task, row_ids = 1:120)
predictions = learner$predict(task, row_ids = 121:150)

For further information and examples, please view the mlr3learners.lightgbm package vignettes and the mlr3book.

GPU acceleration

The mlr3learners.lightgbm can also be used with lightgbm's GPU compiled version.

To install the lightgbm R package with GPU support, execute the following commands (lightgbm manual):

git clone --recursive --branch stable --depth 1 https://github.com/microsoft/LightGBM
cd LightGBM && \
sed -i -e 's/use_gpu <- FALSE/use_gpu <- TRUE/g' R-package/src/install.libs.R && \
Rscript build_r.R

In order to use the GPU acceleration, the parameter device_type = "gpu" (default: "cpu") needs to be set. According to the LightGBM parameter manual, 'it is recommended to use the smaller max_bin (e.g. 63) to get the better speed up'.

learner$param_set$values = mlr3misc::insert_named(
  learner$param_set$values,
  list(
    "objective" = "multiclass",
    "device_type" = "gpu",
    "max_bin" = 63L,
    "early_stopping_round" = 10,
    "learning_rate" = 0.1,
    "seed" = 17L,
    "metric" = "multi_logloss",
    "num_iterations" = 100,
    "num_class" = 3
  )
)

All other steps are similar to the workflow without GPU support.

The GPU support has been tested in a Docker container running on a Linux 19.10 host, Intel i7, 16 GB RAM, an NVIDIA(R) RTX 2060, CUDA(R) 10.2 and nvidia-docker.

More Infos: