[Octave-bug-tracker] [bug #44996] statistics: Appropriate results of gev(c|p)df function for small values of k

[Rafael Laboissiere]

URL:
  <http://savannah.gnu.org/bugs/?44996>

                 Summary: statistics: Appropriate results of gev(c|p)df
function for small values of k
                 Project: GNU Octave
            Submitted by: rlaboiss
            Submitted on: Sat 02 May 2015 10:52:10 AM GMT
                Category: Octave Forge Package
                Severity: 3 - Normal
                Priority: 5 - Normal
              Item Group: Inaccurate Result
                  Status: None
             Assigned to: None
         Originator Name: 
        Originator Email: 
             Open/Closed: Open
         Discussion Lock: Any
                 Release: dev
        Operating System: Any

    _______________________________________________________

Details:

When computing the cdf and pdf of the GEV distributions, in functions gevcdf
and gevpdf, respectively, the parameter k appears in the denominator of a
division operation.  To avoid division by zero, the condition k==0 is tested
and an alternative formula is used for it.

However, this test should be extended to values very close to zero.  Doing
some non-extensive tests, I concluded that the function returns wrong results
for values of abs(k) lesser than 1e-11.  The most appropriate limit value
should be more carefully chosen, though.

This problem was evidenced in unit tests of the gevcdf and gevpdf functions. 
In one case, we have:


k = -0.03 : 0.01 : 0.03;


In my system (Octave 4.0 on Debian i386), the fourth element of k is different
from zero (it is equal 3.4694e-18) and the result of gevcdf is clearly wrong,
and the unit test fails.

Patch attached.



    _______________________________________________________

File Attachments:


-------------------------------------------------------
Date: Sat 02 May 2015 10:52:10 AM GMT  Name: small-k-in-gev-functions.diff 
Size: 2kB   By: rlaboiss

<http://savannah.gnu.org/bugs/download.php?file_id=33876>

    _______________________________________________________

Reply to this item at:

  <http://savannah.gnu.org/bugs/?44996>

_______________________________________________
  Message sent via/by Savannah
  http://savannah.gnu.org/