Path: csiph.com!news.mixmin.net!weretis.net!feeder1.news.weretis.net!news.solani.org!.POSTED!not-for-mail
From: Karl Ratzsch <mail.kfr@gmx.net>
Newsgroups: comp.graphics.apps.gnuplot
Subject: Re: latest change on fit.c
Date: Thu, 5 May 2016 13:40:17 +0200
Organization: solani.org
Lines: 34
Message-ID: <ngfbep$tg6$1@solani.org>
References: <ng5pjn$ran$1@solani.org> <ng6hpe$t8r$1@dont-email.me>
Mime-Version: 1.0
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: 7bit
X-Trace: solani.org 1462448410 30214 eJwFwYEBwCAIA7CXdNDCO1XW/08wQXDzVhJMGGZBNTJbN1ZPGFPfGFi/LlPaliNxsvqUHiuXEcI= (5 May 2016 11:40:10 GMT)
X-Complaints-To: abuse@news.solani.org
NNTP-Posting-Date: Thu, 5 May 2016 11:40:10 +0000 (UTC)
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:45.0) Gecko/20100101 Thunderbird/45.0
X-User-ID: eJwFwYEBwCAIA7CXJtiC51gY/59gAudixSa4MRhGHzUFyS9hMDusNdjsSgUsP8ct/dGR3Q8bbRE+
In-Reply-To: <ng6hpe$t8r$1@dont-email.me>
Cancel-Lock: sha1:Ld+/4zXfi+2EfQQ0lvo3MMJdv2U=
X-NNTP-Posting-Host: eJwFwQERADEIAzBLY1C6yuHh8C/hE3haNiORgcXOd8yvAt4jTOxr0rYOOVJU7+jiqTCWVP4W0RDC
Xref: csiph.com comp.graphics.apps.gnuplot:3323

Am 02.05.2016 um 05:36 schrieb Ethan A Merritt:
> Karl Ratzsch wrote:
> 
>> I'm intrigued what this change
>>
>> http://gnuplot.cvs.sourceforge.net/viewvc/gnuplot/gnuplot/src/fit.c?r1=1.163&r2=1.164
>>
>> does to my fit. Is there a test case to show the difference?
> 
> I do have a test case, but it's real data in the process of real analysis and
> not really in a form I can share.  The long and short of it is that for 
> whatever reason the fit goes bad during iterative refinement and triggers
> an attempt to calculate the log of a negative number. The old code attempted
> to trap such a case and print out diagnotic information by doing this:
> 
>  data[i] = real(&v);	         
>  if (undefined || isnan(real(&v))) {
>       ... print stuff ...
> 
> The problem I ran into was that the call to real(&v) failed immediately
> with error "unknown type in real()" so it never got as far as the
> diagnostic printout.  After the change the test for "undefined" is done
> first, so the diagnostic printout is triggered before any attempt to
> call real(&v).
> 
> Anyhow, the only change is that now you get the intended diagnostic
> printout rather than an "unknown type" error.


Ah. I think i've never run into that one, don't remember ever
fitting a function with a log in it. Thanks for the explanation!

   Karl