URL: <https://savannah.gnu.org/bugs/?66800>
Summary: Feature: gsl_ran_negative_binomial_pdf with p = 1
Group: GNU Scientific Library
Submitter: fermelelundi
Submitted: Sat 15 Feb 2025 10:38:17 AM UTC
Category: Accuracy problem
Severity: 3 - Normal
Operating System:
Status: None
Assigned to: None
Open/Closed: Open
Discussion Lock: Any
Release: 2.8
_______________________________________________________
Follow-up Comments:
-------------------------------------------------------
Date: Sat 15 Feb 2025 10:38:17 AM UTC By: Fermé le Lundi <fermelelundi>
As per Eric Redekopp (University of Saskatchewan), cf
https://lists.gnu.org/archive/html/bug-gsl/2023-09/msg00001.html
gsl_ran_negative_binomial_pdf gives -NaN if p == 1.0, despite p=1 being a
valid negative binomial distribution. I believe this is because of the call to
log1p(-p) which evaluates to log(0) and gives NaN. There appear to be guards
against this in the positive version (gsl_ran_binomial_pdf) and it works as
expected, however these edge cases aren't covered in the negative version. I
think that the negative version should have similar guards to return
appropriate values for edge cases like p==1.
The attached patch fixes this issue and aligns with the analogous fix in
binomial.c.
_______________________________________________________
File Attachments:
-------------------------------------------------------
Name: nbinomial.diff Size: 606B
<https://file.savannah.gnu.org/file/nbinomial.diff?file_id=56899>
AGPL NOTICE
These attachments are served by Savane. You can download the corresponding
source code of Savane at
https://savannah.gnu.org/source/savane-9aa36b8aaaab70f75d4f4abb1d9f8d6bd3eed2b1.tar.gz
_______________________________________________________
Reply to this item at:
<https://savannah.gnu.org/bugs/?66800>
_______________________________________________
Message sent via Savannah
https://savannah.gnu.org/
signature.asc
Description: PGP signature
