This R package provides an interface to the spatialML package for the mlr3 ecosystem.
It implements the Geographically Weighted Random Forest (GRF) algorithm for regression tasks.
Documentation
The site can be found at : https://mlr3learners-spatialml-grf.netlify.app/
This site includes API references, usage guides and a detailed performance benchmark (vignettes).
Installation
To install it, you can use this command :
# install.packages("remotes")
# Installing SpatialML
remotes::install_version("SpatialML", version = "0.1.6")
remotes::install_github("MomoNagi/mlr3learners.spatialML")Usage
This package provides the regr.grf learner for mlr3. It requires a TaskRegrST (Spatio-Temporal Task) from the mlr3spatiotempcv package to correctly handle spatial coordinates.
library(mlr3)
library(mlr3spatiotempcv)
library(mlr3learners.spatialML)
task <- tsk("california_housing")
learner <- lrn("regr.grf", bw = 20, ntree = 50)
learner$train(task)
pred <- learner$predict(task)Benchmark results
Performance was evaluated using the Root Mean Squared Error (RMSE). Results are based on a 5-fold cross-validation. Values represent the Mean RMSE (± Standard Deviation).
| Dataset | GRF | KNN (Tuned) | CV-Glmnet | Featureless |
|---|---|---|---|---|
| California Housing | 91,628 (±8,849) | 88,497 (±9,489) | 81,158 (±10,974) | 123,044 (±8,091) |
| Income | 1,776 (±146) | 1,630 (±290) | 1,921 (±469) | 2,929 (±447) |
Note : Lower values indicate better performance. Bold values represent the best learner for each task.
We can see that GRF remains competitive with KNN and CVGlmnet. It also stands out as the most stable learner on the Income dataset with the lowest standard deviation. Furthermore, its significant outperformance of the Featureless baseline confirms its effectiveness.
Development
This package includes :
- End-to-End Testing: Automated test suite ensuring seamless integration between mlr3 and SpatialML
- Continuous Integration: Validation via GitHub Actions on every push.
- Documentation: Complete API reference and performance vignettes deployed on Netlify.
- Vignette: Complete demonstration of the benchmark
Related work
- Course wiki
- SpatialML (CRAN Archive) - Core package for GRF.
- mlr3spatiotempcv - Required for handling spatial tasks in mlr3.
- Issue mlr3extralearners #403 (SpatialML grf)
Author
Manh Hung LE - GitHub