RoLDSIS – Regression on Low-Dimension Spanned Input Space

Introduction

The present repository contains the data and the code for analysing the data related to a publication involving the RoLDSIS method. RoLDSIS stands for “Regression on Low-Dimension Spanned Input Space”. The data for an electroencephalography (EEG) experiment involving phonemic identification is included in the repository. The scripts in the R language, for processing and analysing the data, as well as producing the figures that appear in the following article:

Santana, A., Barbosa A., Yehia H., and Laboissière R. (2021) A dimension reduction technique applied to regression on high dimension, low sample size neurophysiological data sets. BMC Neurosci 22(1). DOI: 10.1186/s12868-020-00605-0.

Installation

The code can be obtained with the following command:

git clone https://github.com/RoLDSIS/code RoLDSIS

N.B.: Developers with write access right to the repository should do, instead:

git clone [email protected]:RoLDSIS/code.git RoLDSIS

Running the analysis and producing the figures

The system packages R, make, and pdftk are needed. The necessary R packages are installed as required by the scripts.

In GNU/Linux systems, the analysis of the data and the generation of the figures are done with the following commands :

cd RoLDSIS/script
make

Description of the contents of this repository

Scripts containing supporting functions and general parameters definition

• paths.r: Definition of paths for saving figures and results
• dwt-parameters.r: Definition of DWT parameters
• geom-lib.r: Geometry supporting function
• dwt-lib.r: DWT supporting functions
• scalogram.r: Plot the DWT scalogram
• remove-spikes.r: Remove spikes from signals, using DWT
• chisq-to-normal.r: Convert Chi-squre to normal distribution

RoLDSIS functions

• roldsis.r

Cross-validation for RolDSIS, LASSO, Ridge Regression and SPLS

• cross-validation.r: Basic cross-validation function
• compare-methods.r: Regression methods
• run-cv.r: Main script for running the cross-validation

Generation of figures

• figure-cv-scalogram.r
• figure-cv-errors.r
• figures-mean-erp.r
• figures-roldsis-results.r
• figures-id-exp.r
• figure-trials-observation.r

Data

• VOT/Snn.dat: Results from the phonemic identification task
• Snn/cross-validation.dat: Raw EEG data

Description of the generated figures:

• psy-VOT-Snn.pdf: Results of phonemic identification experiment for subject nn
• erp-VOT-Ativo-Snn.pdf: Average ERPs for the five stimuli for subject nn
• cv-direction-Snn.pdf: RoldSIS result represented in scalogram for subject nn
• cv-projections-Snn.pdf: RoLDSIS projected responses for subject nn
• cv-errors.pdf: Cross-validation errors for RoLDSIS, LASSO, Ridge Regression, and SPLS
• cv-scalograms.pdf: Histograms for the regression methods of regressed coefficients
• trials-observation.pdf: RMS prediction error for all subjects using different number of averaged points

Documentation

• doc/RoLDSIS.pdf: Early version of the paper describing the method

Licensing conditions

The files in this repository are made available under the conditions of the GNU Public License, version 3 or later. No warranties. The RolDSIS has been submitted for publication. If you use the software or the data of this repository, please give credit.

Authors

• Copyright © 2020 Rafael Laboissière
• Copyright © 2020 Adrielle de Carvalho Santana
• Copyright © 2020 Hani Camille Yehia