learningmachine is a package for Machine Learning that includes uncertainty quantification for regression and classification (work in progress), and explainability through sensitivity analysis. This is the most stable version of learningmachine for R: the one you should use (update 2024-07-10: use v2.0.1).
Install learningmachine from GitHub (in R console, tested on macOS, Google Colab, and Posit Cloud)
remotes::install_github("Techtonique/learningmachine")
So far, learningmachine offers a unified interface for:
lm: Linear modelbcn: Boosted Configuration ‘neural’ Networks, see https://www.researchgate.net/publication/380760578_Boosted_Configuration_neural_Networks_for_supervised_classificationextratrees: Extremely Randomized Trees; see https://link.springer.com/article/10.1007/s10994-006-6226-1glmnet: Elastic Net Regression; see https://glmnet.stanford.edu/krr: Kernel Ridge Regression; see for example https://www.jstatsoft.org/article/view/v079i03ranger: Random Forest; see https://www.jstatsoft.org/article/view/v077i01ridge: Ridge regression; see https://arxiv.org/pdf/1509.09169xgboost: a scalable tree boosting system see https://arxiv.org/abs/1603.02754svm: Support Vector Machines (only for regression, for now), see https://cran.r-project.org/web/packages/e1071/vignettes/svmdoc.pdf
There are only 2 classes Classifier and Regressor, with methods fit and predict and summary, and all these models can be enhanced by using a quasi-randomized layer that basically augments their capacity. The 3 package vignettes are a great way to get started. Along with the (work in progress, as I’m struggling a little bit with documenting R6 objects) documentation, they’ll eventually be available here:
https://techtonique.r-universe.dev/learningmachine
There are also unit tests in the tests folder on GitHub.
PS: In these slides, in present probabilistic time series forecasting with nnetsauce