Random (in-place) shuffle of input wave's elements

Here is a function that does an in-place permutation of an input wave's elements. I have also included a simple test function that I used to test 'shuffle()'. The code is adapted from several sources on the Web, and uses Igor's 'enoise()' function to generate the random permutation indices.
function shuffle(inwave)    //  in-place random permutation of input wave elements
    wave inwave
    variable N  =   numpnts(inwave)
    variable i, j, emax, temp
    for(i = N; i>1; i-=1)
        emax = i / 2
        j =  floor(emax + enoise(emax))     //  random index
//      emax + enoise(emax) ranges in random value from 0 to 2*emax = i
        temp        = inwave[j]
        inwave[j]       = inwave[i-1]
        inwave[i-1] = temp

function test(npts, nshuffles)
    variable npts, nshuffles
    make/O/N=(nshuffles, npts) wave0 = q+1  //  2D  (trials) x (array size)
    make/O/N=(npts) wtemp                   //  temporary column (trial) wave
    variable i
    for(i=0; i<nshuffles; i+=1)                 //  shuffle each row
        wtemp = wave0[i][p]
        wave0[i][] = wtemp[q]
    Make/O/N=(npts) wmean   //  mean of the trials column for each array 'bin'
    for(i=0; i<npts; i+=1)
        MatrixOp/O wdest = col(wave0,i)
        wmean[i] = mean(wdest)  //  in this test, should approach (npts+1)/2
I think a faster way would be to use the Sort operation

Function Shuffle(InWave)
Wave InWave
Variable n=numpnts(InWave)
Make/o/N=(n) order=enoise(n)
Sort order, InWave
I thought it would be faster too, but it's not. I used startmstimer + stopmstimer, but it turns out using the inbuilt sort operation is a factor of two slower.
Thanks for the suggestion, and the timing test. I was not attracted to the sort method for mathematically pedantic reasons. It is conceivable (although extremely unlikely) that the random (continuous) sorting sequence might have some identical values. The crux of the shuffle algorithm is to change the available integer sample space after each sub-permutation to strictly avoid any kind of duplication problem. I might add that have also inspected the standard deviation of any given element after many trials, and it too agrees with theory. That said, had the sort method been significantly faster, I probably would use it if speed were an issue.




Igor Pro 9

Learn More

Igor XOP Toolkit

Learn More

Igor NIDAQ Tools MX

Learn More