This R package enables you to apply Data-Driven Random Projection and Screening for High-Dimensional Generalized Linear Models (see Parzer, Filzmoser and Vana-Guer 2024a for Linear Regression and Parzer, Filzmoser and Vana-Guer 2024b for the extension to GLMs). The package is presented in more detail in 'spar: Sparse Projected Averaged Regression in R' by Roman Parzer, Laura Vana-Gür and Peter Filzmoser (2024c). Exported functions are
spar: performs the procedure for given thresholds lambda and numbers of marginal models, and acts as a help-function for the full cross-validated procedure spar.cv.spar.cv: performs the full procedure as described in the corresponding paper will cross-validation of the optimal threshold and number of models.
# install.packages("remotes")
remotes::install_github("RomanParzer/SPAR") # for current main branch
# remotes::install_github("RomanParzer/SPAR@*release") # for latest releaseThe method is designed for linear regression using a high-dimensional data set with more variables than observations.
The two main functions return objects, for which coef, predict and plot functions are available.
require(spar)
data("example_data")
spar_res <- spar.cv(example_data$x,example_data$y,nummods=c(5,10,15,20,25,30))
spar_res
coefs <- coef(spar_res)
pred <- predict(spar_res,example_data$x)
plot(spar_res)
plot(spar_res,"MSE","nummod")
plot(spar_res,"numAct","lambda")This package is free and open source software, licensed under GPL-3.