Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: ponder_on ponder_off comparision

Author: Ralf Elvsén

Date: 03:40:38 07/18/00

Go up one level in this thread


On July 17, 2000 at 19:09:20, Mogens Larsen wrote:

>On July 17, 2000 at 17:26:37, Ralf Elvsén wrote:
>
>>We obviously have different opinions on in which settings the most
>>interesting results can be found. I think human opposition or two-comp
>>games are most important. You think one-comp games. They may be more
>>frequent here, but the people posting here are a little special :)
>
>No, we don't disagree on the most interesting results. That is obviously from
>either human-comp or comp-comp with ponder, but that isn't really the issue.
>Most amateur comp-comp testing is about comparative strength, where ponder off
>is equally good as a measurement of that, and there's more data to process due
>to fewer hardware restrictions. You may not like that, but that's the way it is.

Let me make one thing clear: if a programmer explicitely says his program
has a good ponder = OFF management I have few problems with testing the
program in this mode (see below).

>
>>It doesn't take a lot of effort and money to arrange comp-comp games
>>if one really wants to. Human opposition is the ultimate testground
>>for many years to come.
>
>I wouldn't be too sure of that, unless you want to stare at ICC games.

I prefer to stare at Junior at Dortmund.

>
>>Of course ponder OFF is supported and ponder ON is almost equally trivial to
>>implement. The existence of these features, however, has nothing to do
>>with how well they are supported. Time management is a complex affair and
>>a programmer must invest a lot of time to solve it. So with minimal effort
>>both modes can be supported in the sense that they exist. Then comes the
>>tricky part of doing it good. The programmer chooses: do I implement both or
>>one, and if so which one? Maybe some/many of the amateur programmers
>>concentrate on ponder = OFF. One thing must be certain though: the
>>professionals make sure their ponder = ON management is as good as it
>>can be. How much time they spend on ponder = OFF is often completely
>>unknown. And one professional, Bob, doesn't spend any time at all
>>after his rudimentary implementation. He has stated as much that it
>>takes endless hours of testing to get it right and it's not worth the
>>effort. I suspect there are all kinds between the extremes.
>
>I'm not sure you know what you are talking about at all. You have no idea
>about how rudimentary the implementation of ponder OFF of Crafty is
>and how it compares to other programs.

I know how rudimentary it is in the sense that Bob himself says it is
something he never works with. I can't swear he has used the word
"rudimentary" but to say that he doesn't think his ponder = OFF management
is a finely tuned piece of code
is an understatement. How it compares to other programs I don't know and
this is exactly my point. If I want to estimate performance under optimal
conditions (ponder = ON) with results from ponder = OFF this introduces an
uncontrolled parameter.

>I suspect Hyatt's rudimentary implementation is
>quite adequate

So do I. At least it isn't any worse than e.g. Hiarcs' . But I only
look at these games because I have convinced myself of this. I wouldn't
take an unknown program, run 100 ponder = OFF games overnight and then
think the result reflected its playing strength with ponder = ON. I would
look over the games, and in particular the time management with "utter
suspicion". And even if everything seemed OK I couldn't be sure it told me
anything about ponder = ON. Perhaps the programmer hasn't cared about
ponder = ON, but then I consider the program to be unfinished. It's
like it doesn't know about underpromotion or whatever, and I'm not
interested anymore.

>especially since you don't know how the ponder ON implementation
>measure up to the competition either.

All I need to know is that the programmer wants to compete in the
ponder = ON business. Then it is just another aspect of the program,
like king safety or endgame knowledge. Its quality is affecting the
performance of the program, together with every other piece of knowledge
it has, and is reflected in the results. And this is what I care about.

You have a point when you're saying (if I understand you correctly) that
the relative quality of ponder ON/OFF management for two programs is
difficult to know and poses a problem. Even if both programmers says
both modes are competitive, one program can be relatively weak in
ponder = ON and strong in ponder = OFF. Since I think the ponder = ON
mode is the only interesting game, this is another argument for me to
not take ponder = OFF games so seriously without careful observations.
If one thinks ponder ON/OFF is of equal importance,
(which you obviously don't) the conclusion is different but that
would be two different kinds of chess.

Ralf

>
>Best wishes...
>Mogens



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.