[email protected]
[Top] [All Lists]

Re: [Haskell-cafe] Testing for statistical properties

Subject: Re: [Haskell-cafe] Testing for statistical properties
From: Tom Nielsen
Date: Fri, 8 Jan 2010 10:59:11 +0000
Hi Greg,

Assuming this is a one-dimensional distribtution, you should use a
kolmogorov-smirnov test to test this:

http://en.wikipedia.org/wiki/Kolmogorov-Smirnov_test

I've implemented to the KS distribution from the CERN code linked in
the wikipedia article, here:

http://github.com/glutamate/samfun/blob/master/Math/Probably/KS.hs

(warning, i wasn't able to verify the numbers coming out against
anything so just check that it makes sense)

So all you have to do is to find the maximal distance between your
samples and the cumulative density function, multiply by the sqrt. of
of the number of samples, and calculate kprob on that.

I don't think you can do this in a Bayesian way because you can't
enumerate all the other distributions your samples could come from?

Tom

On Thu, Jan 7, 2010 at 9:31 PM, Gregory Crosswhite
<[email protected]> wrote:
> Hey everyone!  I have some computations that satisfy statistical properties 
> which I would like to test --- that is, the result of the computation is 
> non-deterministic, but I want to check that it is sampling the distribution 
> that it should be sampling.  Is anyone aware of a Haskell library out there 
> that does anything like this?
>
> Cheers,
> Greg
>
> _______________________________________________
> Haskell-Cafe mailing list
> [email protected]
> http://www.haskell.org/mailman/listinfo/haskell-cafe
>
_______________________________________________
Haskell-Cafe mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/haskell-cafe

<Prev in Thread] Current Thread [Next in Thread>