https://github.com/scnext/scglr
An extension of the Fisher Scoring Algorithm to combine PLS regression with GLM estimation in the multivariate context. Covariates can be grouped in themes.
https://github.com/scnext/scglr
package partial-least-squares-regression r
Last synced: 8 months ago
JSON representation
An extension of the Fisher Scoring Algorithm to combine PLS regression with GLM estimation in the multivariate context. Covariates can be grouped in themes.
- Host: GitHub
- URL: https://github.com/scnext/scglr
- Owner: SCnext
- Created: 2018-06-25T11:25:08.000Z (over 7 years ago)
- Default Branch: master
- Last Pushed: 2025-03-31T09:27:18.000Z (9 months ago)
- Last Synced: 2025-04-15T22:56:38.594Z (8 months ago)
- Topics: package, partial-least-squares-regression, r
- Language: R
- Homepage: https://scnext.github.io/SCGLR/
- Size: 8.63 MB
- Stars: 2
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.Rmd
- Changelog: NEWS.md
Awesome Lists containing this project
README
---
title: "readme"
author: "X. Bry, G. Cornu, F. Mortier and C. Trottier"
date: "25 juin 2018"
output:
md_document:
variant: gfm
link-citations: true
bibliography: assets/SCGLR.bib
---
# SCGLR 
[](https://cran.r-project.org/package=SCGLR)
[](https://scnext.r-universe.dev/SCGLR)
## Introduction
**SCGLR** is an open source implementation of the Supervised Component Generalized Linear Regression
[@bry13;-@bry16;-@bry18], which identifies, among a large set of potentially multicolinear predictors, the strong dimensions most predictive of a set of responses.
**SCGLR** is an extension of partial least square regression (PLSR) to the uni- and multivariate generalized linear framework. PLSR is particularly well suited for analyzing a large array of explanatory variables and many studies have demonstrated its predictive performance in various biological fields such as genetics [@boulesteix07] or ecology [@carrascal09]. While PLSR is well adapted for continuous variables, maximizing the covariance between linear combination of dependent variables, and linear combinations of covariates, **SCGLR** is suited for non-Gaussian outcomes and non-continuous covariates.
**SCGLR** is a model-based approach that extends PLS [@tenenhaus98], PCA on instrumental variables [@sabatier89], canonical correspondence analysis [@terbraak87], and other related empirical methods, by capturing the trade-off between goodness-of-fit and common structural relevance of explanatory components. The notion of structural relevance has been introduced [@bry15].
**SCGLR** can deal with covariates partitioned in several groups called "themes", plus a group of additional covariates. Each theme is searched for orthogonal components representing its variables in the model, whereas the additional covariates appear directly in the model, without the mediation of a component [@bry18].
**SCGLR** works also for mixed models using an extension of the Schall's algorithm to combine Supervised-Component regression with GLMM estimation in the multivariate context.
## Installation
``` r
# Install release version from CRAN
install.packages("SCGLR")
# Install development version from GitHub
remotes::install_github("SCnext/SCGLR")
# Install development version from R universe
install.packages("SCGLR", repos = c("https://scnext.r-universe.dev", "https://cloud.r-project.org"))
```
## Main functions and works in progress
**SCGLR** is designed to deal with outcomes from multiple distributions: Gaussian, Bernoulli, binomial and Poisson separately or simultaneously [@bry13]. Moreover **SCGLR** is also able to deal with multiple conceptually homogeneous explanatory variable groups [@bry18].
**SCGLR** is a set of **R** functions illustrated on a floristic data set, _genus_. `scglr` and `scglrTheme` are respectively dedicated to fitting the model with one or more thematic group of regressors. `scglrCrossVal` and `scglrThemeBackward` are respectively dedicated to selecting the number of components. `print`, `summary` and `plot` methods are also available for the `scglr` and `scglrTheme` function results.
Different works are in progress both dealing for instance with the inclusion of random effects extending **SCGLR** to the generalized linear mixed model framework [@chauvet18;-@chauvet18b], or the Cox regression model.
## Fundings
The GAMBAS project funded by the Agence Nationale pour la Recherche (ANR-18-CE02-0025).
## References