Supplementary Materials for: “Penalized rank-one approximation to functional partial least-squares regression over complex domains”

Authors: Harold A. Hernández-Roig (haroldantonio.hernandez@uc3m.es); M. Carmen Aguilera-Morillo; Eleonora Arnone; Rosa E. Lillo; Laura M. Sangalli.

Overview

You can download this R Project to reproduce our simulation studies and check the three-dimensional (3D) animations of the estimated coefficient functions relative to the real data application. The repository is organized into three sub-folders:

Folder Animated Betas contains 3D animations of the estimated coefficient functions from Section 5: “Application to brain connectivity data”.
Folders 1D_simulations and 2D_simulations contains the scripts required to reproduce the simulations in Section 4: “Simulation studies”.

Simulation studies

The simulations depend on the R package penR1FPLS. It can be installed as follows:

devtools::install_github("hhroig/penR1FPLS", dependencies = TRUE)

One-dimensional domain

Run:

source("1D_simulations/main_1D.R")

to reproduce the simulations for data defined over a one-dimensional (1D) domain. The script creates a new folder with results, including plots.

Two-dimensional planar domain

Run:

source("2D_simulations/main_2D.R")

to reproduce the simulations for data defined over a two-dimensional (2D) planar domain. The script creates a new folder with results, including plots.

Parallel computing

By default the simulations run in parallel using the doParallel package. For example, in the one-dimensional case, the core script is executed using:

library(doParallel)
nodes_CL = detectCores()   # Detect number of cores to use
cl = makeCluster(nodes_CL) # Specify number of threads here
registerDoParallel(cl)

source("1D_simulations/simulations_1D.R", local = TRUE)

stopCluster(cl)

The simulations can be executed in series by commenting the lines with doParallel commands and sourcing the script as it is: