inst/india_rqss.R

In mboost: Model-Based Boosting

#-----------------------------------------------------------
# Path, Packages, source functions
#-----------------------------------------------------------

library("quantreg")
load("india.Rdata")
source("india_rqss_lambdaOptFunc.R")

#-----------------------------------------------------------
# Quantiles and Indices
#-----------------------------------------------------------

quantiles <- c(0.05, 0.1, 0.5)
inds <- paste("w", 1:50, sep="")


for(i in 1:length(inds)){
    
    indiaTrain <- india[india[,inds[i]]==1,]
    
    indiaTuning <- india[india[,inds[i]]==2,] 
    indiaTuning <- convexHull(dataOrg=indiaTuning, dataRef=indiaTrain)
    
    indiaTest <- india[india[,inds[i]]==3,] 
    indiaTest <- convexHull(dataOrg=indiaTest, dataRef=indiaTrain)
    
    
    for(j in 1:length(quantiles)){
        
        
        # Starting parameters for lambda equal to 100:
        #optimLambda <- optim(rep(100,6), fn=lambdaOpt, fitdata=indiaTrain, 
        #                     evaldata=indiaTuning, tau=quantiles[j], lower=0)                     
        
        # Next try: Starting parameters equal to 10:
        optimLambda <- optim(rep(10,6), fn=lambdaOpt, fitdata=indiaTrain, 
                             evaldata=indiaTuning, tau=quantiles[j], lower=0) 
        lambdaMin <- optimLambda$par
        
        formule <- buildFormula(lambdas=lambdaMin)
        model <- rqss(formule, data=indiaTrain, tau=quantiles[j])
        
        # Results with starting parameters for lambda equal to 100:
        save(model, file=paste("rqssObjects/model",i,"_",quantiles[j],".Rdata",sep=""))
        rm(model)
        gc()
        
    }
}