Hi,
I don't known about the ROOT Kolmogorov-Smirnov method, but,
theoretically, the resulting probability should be uniformly distributed
between 0 and 1 ("flat") for comparisons of distributions randomly
generated from the same parent distribution.
I am "almost sure" that there is an excess near 1 because all
samples have the same size (all histograms have the same number of bins).
Bye,
Sérgio
On Fri, 22 Dec 2000, Andy Salnikov wrote:
>Hi all,
>
>it seems that kolmogorov test does not work for me on weighted histograms.
>Here is the macro:
>
>{
>gROOT->Reset() ;
>
>TH1* h1 = new TH1F("h1","Gauss",200,-5,5);
>TH1* h2 = new TH1F("h2","Gauss",200,-5,5);
>TH1* h3 = new TH1F("h3","Gauss",200,-5,5);
>TH1* h4 = new TH1F("h4","Gauss",200,-5,5);
>
>int j ;
>double xs1, xs2 ;
>
>h1->Reset() ;
>h2->Reset() ;
>
>for ( j=0; j<10000; ++j ) {
> xs1 = gRandom->Gaus(0,1);
> xs2 = gRandom->Gaus(0,1);
> h1->Fill(xs1);
> h2->Fill(xs2);
>}
>
>for ( j = 1 ; j<= 200; ++j ) {
> h3->SetBinContent( j, h1->GetBinContent( j ) / 100 ) ;
> h3->SetBinError( j, h1->GetBinError( j ) / 100 ) ;
>
> h4->SetBinContent( j, h2->GetBinContent( j ) / 100 ) ;
> h4->SetBinError( j, h2->GetBinError( j ) / 100 ) ;
>}
>
>h1->KolmogorovTest(h2,"D") ;
>h3->KolmogorovTest(h4,"D") ;
>
>}
>
>And here its output:
>
> Kolmo Prob h1 = h1, sum1=10000
> Kolmo Prob h2 = h2, sum2=10000
> Kolmo Probabil = 0.366721, Max Dist = 0.013
> Kolmo Prob h1 = h3, sum1=100
> Kolmo Prob h2 = h4, sum2=100
> Kolmo Probabil = 1, Max Dist = 0.013
>
>I'd expect that probabilities be equal for h1/h2 and h3/h4. What gives?
>
>This was obtained in 2.25/03.
>
>Also probably related question. Should the probability from KolmogorovTest
>have flat distibution? When I try call it many times for the above gaussian
>histograms I observe an exess for prob close to 1.
>
> Cheers,
> Andy.
>
>
This archive was generated by hypermail 2b29 : Tue Jan 02 2001 - 11:50:40 MET