Inefficient Non-Parametric Test

[chrstphlbr]

Hi,

I just realized while running both parametric and non-parametric versions of the SK test that the non-parametric version is much slower. I have not measured or profiled the executions, although looking at the code, I suspect the diff function in PartitionNonParametric is the culprit. Once a magnitude is non-negligible, the variable negligible is not updated anymore but the nested loop continues to execute. Unfortunately, I am unable to send a PR myself at the moment, but I can provide a potential fix below.

Current version:

diff <- function(k, g, av, means){
        negligible <- TRUE
        for(i in k:(g-1)){
            for(j in (i+1):g){
                a <- av$model[av$model[,2] == names(means[k]),1]
                b <- av$model[av$model[,2] == names(means[g]),1]   
                magnitude <- as.character(effsize::cliff.delta(a,b)$magnitude, paired=TRUE)
                if(magnitude != "negligible"){
                    negligible <- FALSE
                }
            }
        }
        return(negligible)
    }

Fixed version:

diff <- function(k, g, av, means){
        for(i in k:(g-1)){
            for(j in (i+1):g){
                a <- av$model[av$model[,2] == names(means[k]),1]
                b <- av$model[av$model[,2] == names(means[g]),1]   
                magnitude <- as.character(effsize::cliff.delta(a,b)$magnitude, paired=TRUE)
                if(magnitude != "negligible"){
                    return(FALSE)
                }
            }
        }
        return(TRUE)
    }

Cheers
Christoph