R/misClustRate.R

In norbertbin/SpecClustPack: Spectral Clustering for Network Analysis

#' Calculates the empirical proportion of misclustered nodes.
#'
#' @param clusters A vector of node cluster memberships
#' @param nMembers The number of nodes in each block
#'
#' @export
#' @return The proportion of misclustered nodes up to identifiability.
#'
#' @keywords clustering
misClustRate <- function(clusters, nMembers) {

  nNodes = sum(nMembers)
  nBlocks = length(nMembers)
  nMisClustNodes = 0
  uniqueClusters = unique(clusters)
  clusterCounts = matrix(0, ncol = nBlocks, nrow = nBlocks)
  clusterLabels = rep(0, nBlocks)
  
  #get label counts for each cluster
  for(i in 1:nBlocks) {
    
    clustStart = sum(nMembers[1:i]) - sum(nMembers[i]) + 1
    clustEnd = sum(nMembers[1:i])
    
    for(j in uniqueClusters) {
      clusterCounts[j,i] = sum(j == clusters[clustStart:clustEnd]) 
    }    
  }
  
  #determine cluster label based on counts
  clusterCountsTemp = clusterCounts
  for(i in 1:nBlocks) {
    
    maxCoor = t(which(clusterCountsTemp ==
        max(clusterCountsTemp), arr.ind = T))
    clusterLabels[maxCoor[2]] = maxCoor[1]
    clusterCountsTemp[maxCoor[1], ] = rep(-1, nBlocks)
    clusterCountsTemp[, maxCoor[2]] = rep(-1, nBlocks)
    
  }
  
  for(i in 1:nBlocks) {
    nMisClustNodes = nMisClustNodes + sum(clusterCounts[-clusterLabels[i],i])
  }  
  
  
  return( nMisClustNodes/nNodes )
}