R/removeCorrelatedMetrics.R

In suraj-yathish/MCSlibrary:

#' Function to Remove mitigated metrics
#' 
#' A function that automates the process of deleting the highly correlated metrics after the application of varclus.
#' @param Collect correlated metrics, Defaults to dataset, sw.metrics, threshold of varclus (>0.7)  
#' @keywords sw.metrics, varclus.threshold, output 
#' @export
#' @examples 
#' removeCorrelatedMetrics() 

removeCorrelatedMetrics <- function(dataset, sw.metrics, VarClus.threshold) {
  print('Variable Clustering : START')
  output <- {}
  ran <- {}
  correlated <- {}
  mitigated.metrics <- {}
  final.mitigated.metrics <- {}
    
  vc <-
    varclus(as.formula(paste('~',paste(sw.metrics, collapse = "+"))),
            similarity = 'spearman',
            data = dataset,
            trans = "abs", na.action = na.omit)
  
  # Apply cutoff threshold
  var.clustered <-
    cutree(vc$hclust, h = (1 - VarClus.threshold))
  
  # Get surviving metrics
  non.correlated.metrics <-
    names(var.clustered)[var.clustered %in% names(table(var.clustered)[table(var.clustered) == 1])]
  print(paste0(
    length(non.correlated.metrics),
    ' non-correlated metrics : ',
    paste0(non.correlated.metrics, collapse = ', ')
  ))
  
  
  
  # Get correlated clusters index
  correlated.clusters.index <-
    names(table(var.clustered)[table(var.clustered) > 1])
  
  print(paste0((length(sw.metrics) - length(non.correlated.metrics)),
               ' correlated metrics from ',
               length(correlated.clusters.index),
               ' clusters'
  ))
  
  
  

  # For each cluster of correlated metrics, print out for manual selection
  cluster.count <- 1
  for (cluster.index in correlated.clusters.index) {
    print(paste0(
      'Cluster ',
      cluster.count,
      ' : ',
      paste0(names(var.clustered)[var.clustered == cluster.index], collapse = ', ')
    ))
    output <- rbind(output,
                    data.frame(
                      CID = cluster.count,
                      CorrelatedMetrics = paste0(names(var.clustered)[var.clustered == cluster.index], collapse = ',')
                    ))
    correlated <- c(correlated, names(var.clustered)[var.clustered == cluster.index])
    cluster.count <- cluster.count + 1; 
  } 
  
  if(length(sw.metrics) - length(non.correlated.metrics) == 0)
  {
    mitigated.metrics <- non.correlated.metrics
    print(mitigated.metrics)
    return (list (MM = mitigated.metrics))
  } 
     RandSelCorr <- strsplit(names(table(output$CorrelatedMetrics)[table(output$CorrelatedMetrics) == 1]), ",")
    for(i in 1:length(RandSelCorr)) 
      {
        ran[i] <- sample(RandSelCorr[[i]],1,replace = TRUE)
      } 
        print('Random Selection of highly correlated metrics from each Clusters')
        ran <- as.character(ran)
        print(ran)
        mitigated.metrics <- c(non.correlated.metrics, ran)
        mitigated.metrics <- as.character(mitigated.metrics)
        print('Collection of software metrics after 1st iteration of Varclus')
        print(mitigated.metrics)
        print('Variable Clustering : END')
        
        if(length(sw.metrics) - length(non.correlated.metrics) != 0)
        {
          sw.metrics <- mitigated.metrics
          final.mitigated.metrics <- removeCorrelatedMetrics(dataset, mitigated.metrics, 0.7)
        }
       
       
  return (list(correlated.cluster = output,
                 correlated.metrics = correlated,
                 randomselection = ran,
                 MM = mitigated.metrics))
  
}