sparsevinereg: High-dimensional sparse vine copula regression

An R package that selects variables from high-dimensional continuous data to make vine copula based (quantile) univariate predictions.

It depends on vinereg and kde1d.

Installation

You can install the development version from GitHub with:

# install.packages("remotes")
remotes::install_github("oezgesahin/sparsevinereg")

Package overview

Below is an overview of some functions and features.

Simulate data with 5 relevant and 5 irrelevant variables

library(sparsevinereg)
set.seed(11)
x <- matrix(rnorm(5000), 500, 10)
y <- x[,1] -2*x[,2] + 3*x[,3] + 5*x[,4] - 4*x[,5]

# response is in the first column of the data
data <- data.frame(y = y, x = x)

fit a sparse vine copula based regression model

fit <- sparsevinereg(data)

fit a sparse vine copula based regression model based on partial correlations

fit_ParCor <- sparsevinereg(data, varsel='ParCor')

print the main statistics of the model

print(fit)
#> method = resid   vine = Dvine   vars_indx = 4 5 3 2 1

make predictions with the model at the quantile level 0.90

upper_pred <- predict(fit, data, 0.90)

make predictions with the model at the mean

mean_pred <- predict(fit, data, NA)

Contact

Please contact ozgesahin-94@hotmail.com if you have any questions.

References

Sahin, O., and Czado, C. (2022). High-dimensional sparse vine copula regression with application to genomic prediction. arXiv preprint arXiv:2208.12383. preprint.