In robingenuer/VSURF: Variable Selection Using Random Forests

knitr::opts_chunk$set(cache = FALSE)

resFiles <- list.files("outputs/")
for (fileName in resFiles) {
  load(paste0("outputs/", fileName))
}

library(VSURF)

toys data

data(toys)

Sequential with randomForest

set.seed(556264)
toysRF <- VSURF(toys$x, toys$y)

summary(toysRF)
save(toysRF, file = "outputs/toysRF.Rdata")

Sequential with ranger

set.seed(475073)
toysRg <- VSURF(toys$x, toys$y, RFimplem = "ranger")

summary(toysRg)
save(toysRg, file = "outputs/toysRg.Rdata")

Parallel with randomForest

ncores <- 24

set.seed(364693, kind = "L'Ecuyer-CMRG")
toysRFPara <- VSURF(toys$x, toys$y, parallel = TRUE, clusterType = "FORK", ncores = ncores)

summary(toysRFPara)
save(toysRFPara, file = "outputs/toysRFPara.Rdata")

Parallel with ranger

Fork

set.seed(850162, kind = "L'Ecuyer-CMRG")
toysRgPara <- VSURF(toys$x, toys$y, RFimplem = "ranger", parallel = TRUE, clusterType = "FORK", ncores = ncores)

summary(toysRgPara)
save(toysRgPara, file = "outputs/toysRgPara.Rdata")

Internal

set.seed(289817)
toysRgParaInter <- VSURF(toys$x, toys$y, RFimplem = "ranger", parallel = TRUE, clusterType = "ranger", ncores = ncores)

summary(toysRgParaInter)
save(toysRgParaInter, file = "outputs/toysRgParaInter.Rdata")

spam data

data("spam", package = "kernlab")
set.seed(9146301)
levels(spam$type) <- c("ok", "spam")
yTable <- table(spam$type)
indApp <- c(sample(1:yTable[2], yTable[2]/2),
            sample((yTable[2] + 1):nrow(spam), yTable[1]/2))
spamApp <- spam[indApp, ]
spamTest <- spam[-indApp, ]

ncores <- 32

Threshold step

spamThres <- VSURF_thres(type~., spamApp, parallel = TRUE, ncores = ncores, clusterType="FORK")
save(spamThres, file = "outputs/spamThres.Rdata")

spamThres$comput.time

spamRgThres <- VSURF_thres(type~., spamApp, parallel = TRUE, ncores = ncores, clusterType="FORK", RFimplem = "ranger")
save(spamRgThres, file = "outputs/spamRgThres.Rdata")

spamRgThres$comput.time

spamRgThresParaRg <- VSURF_thres(type~., spamApp, parallel = TRUE, ncores = ncores, clusterType = "ranger", RFimplem = "ranger")
save(spamRgThresParaRg, file = "outputs/spamRgThresParaRg.Rdata")

spamRgThresParaRg$comput.time

Interpretation step

spamInterp <- VSURF_interp(type~., spamApp, vars = spamThres$varselect.thres, parallel = TRUE, ncores = ncores, clusterType = "FORK")
save(spamInterp, file = "outputs/spamInterp.Rdata")

spamInterp$comput.time

spamRgInterp <- VSURF_interp(type~., spamApp, vars = spamThres$varselect.thres, parallel = TRUE, ncores = ncores, clusterType = "FORK", RFimplem = "ranger")
save(spamRgInterp, file = "outputs/spamRgInterp.Rdata")

spamRgInterp$comput.time

spamRgInterpParaRg <- VSURF_interp(type~., spamApp, vars = spamThres$varselect.thres, parallel = TRUE, ncores = ncores, clusterType = "ranger", RFimplem = "ranger")
save(spamRgInterpParaRg, file = "outputs/spamRgInterpParaRg.Rdata")

spamRgInterpParaRg$comput.time

spamRbInterp <- VSURF_interp(type~., spamApp, vars = spamThres$varselect.thres, parallel = TRUE, ncores = ncores, clusterType = "FORK", RFimplem = "Rborist")
save(spamRbInterp, file = "outputs/spamRbInterp.Rdata")

spamRbInterp$comput.time

spamRbInterpParaRb <- VSURF_interp(type~., spamApp, vars = spamThres$varselect.thres, parallel = TRUE, ncores = ncores, clusterType = "Rborist", RFimplem = "Rborist")
save(spamRbInterpParaRb, file = "outputs/spamRbInterpParaRb.Rdata")

spamRbInterpParaRb$comput.time

Prediction step

spamPred <- VSURF_pred(type~., spamApp, err.interp = spamInterp$err.interp, varselect.interp = spamInterp$varselect.interp)
save(spamPred, file = "outputs/spamPred.Rdata")

spamPred$comput.time

spamRgPred <- VSURF_pred(type~., spamApp, err.interp = spamInterp$err.interp, varselect.interp = spamInterp$varselect.interp, RFimplem = "ranger")
save(spamRgPred, file = "outputs/spamRgPred.Rdata")

spamRgPred$comput.time

spamRgPredParaRg <- VSURF_pred(type~., spamApp, err.interp = spamInterp$err.interp, varselect.interp = spamInterp$varselect.interp, RFimplem = "ranger", num.threads = ncores)
save(spamRgPredParaRg, file = "outputs/spamRgPredParaRg.Rdata")

spamRgPredParaRg$comput.time

spamRbPred <- VSURF_pred(type~., spamApp, err.interp = spamInterp$err.interp, varselect.interp = spamInterp$varselect.interp, RFimplem = "Rborist")
save(spamRbPred, file = "outputs/spamRbPred.Rdata")

spamRbPred$comput.time

spamRbPredParaRb <- VSURF_pred(type~., spamApp, err.interp = spamInterp$err.interp, varselect.interp = spamInterp$varselect.interp, RFimplem = "Rborist", nThread = ncores)
save(spamRbPredParaRb, file = "outputs/spamRbPredParaRb.Rdata")

spamRbPredParaRb$comput.time

vac18 data

data("vac18", package = "mixOmics")
geneExpr <- vac18$genes
stimu <- vac18$stimulation
ncores <- 3

Parallel with randomForest

vacRF <- VSURF(x = geneExpr, y = stimu, parallel = TRUE, ncores = ncores, clusterType = "FORK")

summary(vacRF)
save(vacRF, file = "outputs/vacRF.Rdata")

Parallel with ranger

vacRgThres <- VSURF_thres(x = geneExpr, y = stimu, parallel = TRUE, ncores = ncores, RFimplem = "ranger", clusterType = "FORK")

vacRgThres$comput.time
save(vacRgThres, file = "outputs/vacRgThres.Rdata")

vacRgInterp <- VSURF_interp(x = geneExpr, y = stimu, parallel = TRUE, ncores = ncores, RFimplem = "ranger", clusterType = "FORK", vars = vacRF$varselect.thres)

vacRgInterp$comput.time
save(vacRgInterp, file = "outputs/vacRgInterp.Rdata")

vacRgPred <- VSURF_pred(x = geneExpr, y = stimu, RFimplem = "ranger", varselect.interp = vacRF$varselect.interp, err.interp = vacRF$err.interp)

vacRgPred$comput.time
save(vacRgPred, file = "outputs/vacRgPred.Rdata")

robingenuer/VSURF documentation built on July 15, 2024, 8:18 p.m.