Multivariate methods with Unbiased Variable selection in
R
PhD candidate Yingxiao Yan
yingxiao@chalmers.se
Associate Professor Carl Brunius
carl.brunius@chalmers.se
Department of Life Sciences, Chalmers University of Technology
www.chalmers.se
The MUVR package allows for predictive multivariate modelling with minimally biased variable selection incorporated into a repeated double cross-validation framework. The MUVR procedure simultaneously produces both minimal-optimal and all-relevant variable selections.
The MUVR2 package is developed with new functionalities based on the MUVR package.
An easy-to-follow tutorial on how to use the MUVR2 package can be found at this repository at inst/Tutorial/MUVR_Tutorial.docx
In brief, MUVR2 proved the following functionality: - Types: classification, regression and multilevel. - Model cores: PLS, Random Forest, Elastic Net. - Validation: repeated double cross-validation (rdCV; Westerhuis et al. 2008, Filzmoser et al. 2009). - Variable selection: recursive feature elimination embedded in the rdCV loop. - Resampling tests and permutation tests: assessment of modelling fitnness and overfitting.
You also need to have the remotes
R package installed.
Just run the following from an R script or type it directly at the R
console (normally the lower left window in RStudio):
install.packages('remotes')
When remotes
is installed, you can install the
MUVR2
package by running:
library(remotes)
install_github('MetaboComp/MUVR2')