Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Why isn't the ability to handicap a program included in current surv

Author: Don Dailey

Date: 21:51:39 06/20/98

Go up one level in this thread


On June 20, 1998 at 23:59:10, Robert Hyatt wrote:

>On June 20, 1998 at 14:11:22, Don Dailey wrote:
>
>>On June 20, 1998 at 13:08:55, Robert Pawlak wrote:
>>
>>>Subject line says it all. With engines being as strong as they are today, I
>>>believe that the ability to dumb-down your software is very important.
>>>
>>>Perhaps the question about playing styles was supposed to encompass this
>>>ability? In either case, this might be an interesting question for future polls.
>>
>>
>>I thought every program had settable levels?  In what way do you want
>>it to be handicapped?
>>
>>- Don
>
>
>levels don't cut it.  I watched someone running a crafty last night
>playing with st=.01, and it was running at 2300 for several dozen games
>without thinking on  opponent's time.  That's still *way* too strong for
>most players in the world, because on a good machine that is still several
>plies.
>
>It's a common complaint...


In Socrates I had a feature that was very popular with people.  They
could set the level by rating.  Basically me and Larry worked out
a formula so that the opponent could ask for 1200 USCF and Socrates
would figure out about how many nodes to search.  It was independent
of hardware too, so 1200 was 1200.  I don't remember what the lowest
level was but I think we tried to make it about the same as a 1 ply
search.  Thinking on the opponents time was not used at these levels,
or if it was the search  always stopped at the estimated level.

Another common technique is to have a very restricted quies search.
I think some of the Novag machines had this.  At the very lowest
level you might have a 1 ply search with 1 or zero ply of quies
search.

You could also have a dumb evaluation function.  I think it's
hard to dumb it down too much without is seeming very artificial.
Probably by eliminating a lot of the evaluation AND restricting
a 1 ply search could give you a program that acted like a real
ameteur.   A lot of programs play quite good with a real 1 ply
search, and would easily beat beginners and advanced beginners
who typically drop pieces fairly often.

- Don



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.