AdaptiveReg: Regression predictor based on an adaptive handling of...

In fhebert/AdaptivePrediction: Adaptive Prediction Rules

Description

This function constructs a predictor for regression problems using an adaptive handling of dependence. Dependence handling is controlled by the number of eigenvectors of the correlation matrix of the explanatory variables. The number of eigenvectors is selected using a cross-validation procedure maximizing the R^2.

Usage

AdaptiveReg(x, y, nvmax = NULL, nfolds = 10)

Arguments

x	A matrix of size n x p containing the observations of the explanatory variables
y	A real-valued vector of length n containing the observations of the response
nvmax	An integer giving the maximum number of latent factors to take to construct the prediction rule. By default (NULL), the function takes as many as possible.
nfolds	The number of folds to use to perform the cross-validation for the selection of the number of eigenvectors. Must be between 3 and n (for leave-one-out cross-validation). Default = 10.

Value

A list containing the following elements:

ZgMoments	A list containing several elements related to the moments of the latent variables involved in the calculation of the predictor
nv	The optimal number of eigenvectors
R2	The vector of values of the cross-validation R^2 (one value for each possible number of eigenvectors).

Examples

library(prospectr)
data(NIRsoil)
indNA = which(is.na(NIRsoil$Nt))
NIRNt = list(x=NIRsoil$spc[-indNA,],y=NIRsoil$Nt[-indNA])
fit = AdaptiveReg(NIRNt$x,NIRNt$y,nvmax=100) #100 latent factors at most

fhebert/AdaptivePrediction documentation built on Nov. 4, 2019, 12:40 p.m.