script.r

# Load libraries

library(foreign)
library(cvTools)

library(randomForest)
library(rpart)
library(e1071)
library(nnet)

# Informational columns

infoCols = c(
  "methodName",
  "commitId"
)

# Uninteresting columns to be discarded

discardCols = c(
  "ADDING_ATTRIBUTE_MODIFIABILITY",
  "ADDING_CLASS_DERIVABILITY",
  "ADDITIONAL_CLASS",
  "ADDITIONAL_FUNCTIONALITY",
  "ADDITIONAL_OBJECT_STATE",
  "ATTRIBUTE_RENAMING",
  "ATTRIBUTE_TYPE_CHANGE",
  "CLASS_RENAMING",
  "COMMENT_DELETE",
  "COMMENT_INSERT",
  "COMMENT_MOVE",
  "COMMENT_UPDATE",
  "DOC_DELETE",
  "DOC_INSERT",
  "DOC_UPDATE",
  "PARENT_CLASS_CHANGE",
  "PARENT_CLASS_DELETE",
  "PARENT_CLASS_INSERT",
  "PARENT_INTERFACE_CHANGE",
  "PARENT_INTERFACE_DELETE",
  "PARENT_INTERFACE_INSERT",
  "REMOVED_CLASS",
  "REMOVED_FUNCTIONALITY",
  "REMOVED_OBJECT_STATE",
  "REMOVING_ATTRIBUTE_MODIFIABILITY",
  "REMOVING_CLASS_DERIVABILITY",
  "UNCLASSIFIED_CHANGE"
)

# Change columns to be used as raw features

rawChangeCols = c(
  "ALTERNATIVE_PART_DELETE",
  "ALTERNATIVE_PART_INSERT",
  "CONDITION_EXPRESSION_CHANGE",
  "STATEMENT_DELETE",
  "STATEMENT_INSERT",
  "STATEMENT_ORDERING_CHANGE",
  "STATEMENT_PARENT_CHANGE",
  "STATEMENT_UPDATE"
)

# Parameter change columns

paramChangeCols = c(
  "PARAMETER_DELETE",
  "PARAMETER_INSERT",
  "PARAMETER_ORDERING_CHANGE",
  "PARAMETER_RENAMING",
  "PARAMETER_TYPE_CHANGE"
)

# Header change columns

headerChangeCols = c(
  "ADDING_METHOD_OVERRIDABILITY",
  "DECREASING_ACCESSIBILITY_CHANGE",
  "INCREASING_ACCESSIBILITY_CHANGE",
  "METHOD_RENAMING",
  "REMOVING_METHOD_OVERRIDABILITY",
  "RETURN_TYPE_CHANGE",
  "RETURN_TYPE_DELETE",
  "RETURN_TYPE_INSERT"
)

# Additional features

additionalFeatureCols = c(
  "numCommits",
  "numAuthors",
  "avgChanges",
  "avgEntities",
  "avgAuthorCommits",
  "avgAuthorChanges",
  "avgChangeRatio",
  "changeGini",
  "timeSinceLastFix"
)

# Bug-proneness columns

bugPronenessCols = c(
  "linBugProneness0.0",
  "geomBugProneness0.5",
  "weightBugProneness"
)

# A function for reading data, preprocessing features 
# and filtering instances

readTrainData = function(filePath) {
  data = read.arff(filePath)
  
  bugProneness = data[bugPronenessCols]
  rawChanges = data[rawChangeCols]
  paramsChanges = apply(data[paramChangeCols], 1, sum)
  headerChanges = apply(data[headerChangeCols], 1, sum)
  additionalFeatures = data[additionalFeatureCols[
    additionalFeatureCols %in% colnames(data)]]
  
  trainData = cbind(
    rawChanges, additionalFeatures, bugProneness,
    paramsChange = paramsChanges, headerChange = headerChanges
  )
  trainData[complete.cases(bugProneness),]
}

# Functions for performing cross-validation of various models

performCV = function(trainData, classColName, ...) {
  x = trainData[, !names(trainData) %in% bugPronenessCols]
  y = trainData[, classColName]
  trainData = cbind(x, class=y)
  complete = complete.cases(trainData)
  
  forestCV = suppressWarnings(
    cvFit(randomForest, class~., trainData, ...))$cv
  treeCV = suppressWarnings(
    cvFit(rpart, class~., trainData, ...))$cv
  svmCV = suppressWarnings(
    cvFit(svm, class~., trainData, ...))$cv
  nnetCV = suppressWarnings(
    cvFit(nnet, x=x[complete,], y=y[complete],
          args=c(size=10, linout=T, trace=F), ...))$cv
  
  c(forest=forestCV, tree=treeCV, svm=svmCV, nnet=nnetCV)
}

performAllCV = function(trainData, K=20, R=10, seed=12345, ...) {
  sapply(bugPronenessCols, function(col) performCV(
    trainData, classColName = col, K=K, R=R, ...
  ))
}

# Read data extracted from repositories

dataSets = list(
  commonsLang = readTrainData("commons-lang.arff"),
  guava = readTrainData("guava.arff"),
  hibernate = readTrainData("hibernate.arff"),
  jetty = readTrainData("jetty.arff"),
  jgit = readTrainData("jgit.arff"),
  junit = readTrainData("junit.arff"),
  log4j = readTrainData("log4j.arff"),
  maven = readTrainData("maven.arff"),
  mockito = readTrainData("mockito.arff"),
  spring = readTrainData("spring.arff")
)
sizes = sapply(dataSets, nrow)

# Perform experiments and aggregate the results

results = lapply(dataSets, performAllCV)
avgResults = apply(simplify2array(results), c(1,2), mean)