[LNT] r207898 - Use Mann-Whitney U test to identify changes

[Hal Finkel]

----- Original Message -----
> From: "Anton Korobeynikov" <anton at korobeynikov.info>
> To: "Yi Kong" <Yi.Kong at arm.com>
> Cc: "LLVM Commits" <llvm-commits at cs.uiuc.edu>
> Sent: Sunday, May 4, 2014 12:31:04 PM
> Subject: Re: [LNT] r207898 - Use Mann-Whitney U test to identify changes
> 
> > In the docstring, it says "Use only when the n in each condition is
> > < 20 and
> > you have 2 independent samples of ranks."
> And this is definitely wrong, since it does use the normal
> approximation, which is valid for n > 20 in each sample ;)

This does indeed seem to be the case; the source code clearly uses the normal approximation. I think that we should just code this ourselves, the algorithm for computing the exact value of U is pretty simple, and a table of critical values for U can be included without much trouble for the few p values we'll care about.

 -Hal

> 
> --
> With best regards, Anton Korobeynikov
> Faculty of Mathematics and Mechanics, Saint Petersburg State
> University
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 

-- 
Hal Finkel
Assistant Computational Scientist
Leadership Computing Facility
Argonne National Laboratory