# HG changeset patch
# User deepakjadmin
# Date 1458723209 14400
# Node ID 69b8598d933832652199e699ba099e4e663f7361
# Parent 16c9aaf658e68ebc3d09ac20baef6680d225c66d
Uploaded
diff -r 16c9aaf658e6 -r 69b8598d9338 feature_selection.R
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/feature_selection.R Wed Mar 23 04:53:29 2016 -0400
@@ -0,0 +1,123 @@
+args <- commandArgs(T)
+
+arg1 <- args[1]
+arg2 <- args[2]
+arg3 <- args[3]
+arg4 <- args[4]
+arg5 <- args[5]
+arg6 <- args[6]
+arg7 <- args[7]
+arg8 <- args[8]
+
+library(caret)
+load(arg1)
+
+RAWDATA <- dataX
+RAWDATA$outcome <- dataY
+rawData <- dataX
+predictorNames <- names(rawData)
+
+isNum <- apply(rawData[,predictorNames, drop = FALSE], 2, is.numeric)
+if(any(!isNum)) stop("all predictors in rawData should be numeric")
+
+colRate <- apply(rawData[, predictorNames, drop = FALSE],
+ 2, function(x) mean(is.na(x)))
+colExclude <- colRate > 0.01
+ if(any(colExclude)){
+ predictorNames <- predictorNames[-which(colExclude)]
+ rawData <- RAWDATA[, c(predictorNames,"outcome")]
+ } else {
+ rawData <- RAWDATA
+ }
+ rowRate <- apply(rawData[, predictorNames, drop = FALSE],
+ 1, function(x) mean(is.na(x)))
+
+rowno <- dim(rawData)[1]
+if (rowno <= 1000){
+cutoff <- rowno / (rowno * 100)
+} else if (rowno > 1000 & rowno <= 5000) {
+cutoff <- rowno / (rowno * 100 * 0.5 )
+} else {
+cutoff <- rowno / (rowno * 100 * 0.5 * 0.5)
+}
+rowExclude <- rowRate > cutoff
+ if(any(rowExclude)){
+ rawData <- rawData[!rowExclude, ]
+ ##hasMissing <- apply(rawData[, predictorNames, drop = FALSE],
+ ##1, function(x) mean(is.na(x)))
+
+############################################################################
+
+
+###############################################################################
+ } else {
+ rawData <- rawData[complete.cases(rawData),]
+
+ }
+
+set.seed(2)
+
+#print(dim(dataX))
+#print(dim(rawData))
+#print(length(dataY))
+
+nzv <- nearZeroVar(rawData[,1:(length(rawData) - 1)])
+ if(length(nzv) > 0) {
+ #nzvVars <- names(rawData)[nzv]
+ rawData <- rawData[,-nzv]
+ #rawData$outcome <- dataY
+ }
+
+predictorNames <- names(rawData)[names(rawData) != "outcome"]
+
+dx <- rawData[,1:length(rawData)-1]
+dy <- rawData[,length(rawData)]
+corrThresh <- as.numeric(arg8)
+highCorr <- findCorrelation(cor(dx, use = "pairwise.complete.obs"),corrThresh)
+dx <- dx[, -highCorr]
+subsets <- seq(1,length(dx),by=5)
+normalization <- preProcess(dx)
+dx <- predict(normalization, dx)
+dx <- as.data.frame(dx)
+
+if (arg4 == "lmFuncs"){
+ctrl1 <- rfeControl(functions = lmFuncs,
+ method = arg5 ,
+ repeats = as.numeric(arg6),
+ number = as.numeric(arg7),
+ verbose = FALSE)
+} else if(arg4 == "rfFuncs"){
+ctrl1 <- rfeControl(functions = rfFuncs,
+ method = arg5 ,
+ repeats = as.numeric(arg6),
+ number = as.numeric(arg7),
+ verbose = FALSE)
+}else if (arg4 == "treebagFuncs"){
+ctrl1 <- rfeControl(functions = treebagFuncs,
+ method = arg5 ,
+ repeats = as.numeric(arg6),
+ number = as.numeric(arg7),
+ verbose = FALSE)
+}else {
+
+ctrl1 <- rfeControl(functions = nbFuncs,
+ method = arg5 ,
+ repeats = as.numeric(arg6),
+ number = as.numeric(arg7),
+ verbose = FALSE)
+}
+
+
+
+
+Profile <- rfe(dx, dy,sizes = subsets,rfeControl = ctrl1)
+
+pred11 <- predictors(Profile)
+save(Profile,file=arg2)
+dataX <- rawData[,pred11]
+dataY <- rawData$outcome
+
+save(dataX,dataY,file=arg3)
+rm(dataX)
+rm(dataY)
+
diff -r 16c9aaf658e6 -r 69b8598d9338 featureselect.zip
Binary file featureselect.zip has changed
diff -r 16c9aaf658e6 -r 69b8598d9338 tool_dependencies.xml
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/tool_dependencies.xml Wed Mar 23 04:53:29 2016 -0400
@@ -0,0 +1,13 @@
+
+
+
+
+ $REPOSITORY_INSTALL_DIR
+
+
+
+
+
+
+
+
diff -r 16c9aaf658e6 -r 69b8598d9338 toolrfe.xml
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/toolrfe.xml Wed Mar 23 04:53:29 2016 -0400
@@ -0,0 +1,87 @@
+
+
+ This tool used for extract best feature subsets cantaining input data for model building.
+
+
+
+ FEATURE_SELECTION_R
+ R_ROOT_DIR
+ R
+ caret-tools
+
+feature_selection.R $input $profile $finalset $function1 $resampling $repeat $number $corcutoff > /dev/null 2>&1
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+.. class:: infomark
+
+**RFE based feature selection for classification and regression**
+
+Input file must be RData file obtained by converting csv file in to RData.
+
+output "Selected_feature.RData" file used for model building purpose.While profile
+
+represents feature selection model.
+
+Correlation cutoff value is desired for choosing independent variables For example
+
+Cutoff value = 0.8 removes all descriptors sharing equal or highet correlation values.
+
+User may choose varous resampling methods in combination with repeats and times of resample.
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+