TreEvo for-loop optimization example 1: before and after

gistfile1.r

##
# before for-loop optimization
##
getSimulationSplits<-function(phy) {
  phy$node.label<-NULL
  branchingTimes<-sort(branching.times(phy), decreasing=TRUE)
  branchingTimesNames<-names(branchingTimes)
  ancestorVector<-c()
  descendant1Vector<-c()
  descendant2Vector<-c()
  for (i in 1:length(branchingTimes)) {
    relationshipVector<-phy$edge[phy$edge[, 1]==branchingTimesNames[i]]
    ancestorVector<-c(ancestorVector, relationshipVector[2])
    descendant1Vector<-c(descendant1Vector, relationshipVector[3])
    descendant2Vector<-c(descendant2Vector, relationshipVector[4])
  }
  return(data.frame(branchingTimes, ancestorVector, descendant1Vector, descendant2Vector))
}

##
# after converting for-loop into vector programming
##
getSimulationSplits<-function(phy) {
  phy$node.label<-NULL
  branchingTimes<-sort(branching.times(phy), decreasing=TRUE)
  branchingTimesNames<-names(branchingTimes)
  # To optimize away the for() loop without parallelization: The basic
  # idea is that each iteration of the loop creates a vector of size
  # 4, of which the last 3 elements get each appended to 3 different
  # lists. Afterwards, the 3 lists get combined together to a
  # datamatrix. Thus, if we can in one step create a matrix consisting
  # of all these 4-element vectors, we can dispose of the for()
  # loop. sapply() does just that:
  m <- sapply(branchingTimesNames, function(i) phy$edge[phy$edge[,1]==i]);
  # remove the first element (only the last 3 are needed), then transpose
  m <- t(m[-1,])
  # compose the dataframe (the naming of the columns may need to be corrected)
  return(data.frame(cbind(branchingTimes, m)));
}