Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!rochester!udel!gatech!howland.reston.ans.net!pipex!oleane!jussieu.fr!univ-lyon1.fr!swidir.switch.ch!newsfeed.ACO.net!edvz.sbg.ac.at!news
From: helmut@cosy.sbg.ac.at (Helmut Mayer)
Subject: Re: Random Number Generator Effects
Message-ID: <D8IJBx.DpC@wst.edvz.sbg.ac.at>
Sender: news@wst.edvz.sbg.ac.at (USENET News System)
Reply-To: helmut@cosy.sbg.ac.at
Organization: University of Salzburg / Austria
References: <3ot2rk$3fq@usenet.srv.cis.pitt.edu>
Date: Sat, 13 May 1995 10:47:57 GMT
Lines: 32

David M. Tate writes

# I've thought about this issue, and I've concluded that it doesn't matter
# what generator you use, as long as it's not *extremely* stupid.

That's my opinion, too. I've done some experiments (with admittedly very simple  
test functions like the "famous" f(x) = x^10) with different PRNGs. There was  
really no difference in on- and off-line performance. Only with extremely  
stupid PRNGs (the ones I built..;-) a significant change in performance could  
be observed. This doesn't even mean that the performance decreased. Due to  
coding inherent effects performance sometimes even increased with a stupid  
PRNG.
However, it should be noted that some people in this group have occasionally  
reported that the use of the standard C-function random() had some performance  
decreasing effects, but I couldn't observe this behavior.
Finally, it should be noted that what constitutes a "good" PRNG is a difficult  
question which is, of course, problem-dependent (how familiar..;-). Measures  
like discrepancy of a PRNG are of mathematical interest, but it is very  
difficult to set up guidelines how to choose a PRNG for a special class of  
problems.

Best regards,

Helmut
---
Helmut A. Mayer, University of Salzburg, Department of Computer Science
*** Member of siGis ***
Jakob-Haringer-Strasse 2, A-5020 Salzburg, AUSTRIA
Tel.: +43-662-8044-6315  Fax: +43-662-8044-611
"...All I want is PEACE and LOVE on this planet, aint how that GOD planned  
it?..." - Chuck D.
 
