Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Time to implement Learning

Author: Bob Durrett

Date: 15:09:33 02/26/04

Go up one level in this thread


On February 26, 2004 at 17:51:52, Tom Likens wrote:

>On February 26, 2004 at 17:35:51, martin fierz wrote:
>
>>On February 26, 2004 at 17:19:20, Tom Likens wrote:
>>
>>>
>>>Well, it's time I broke down and implemented learning.  I've noticed that
>>>my program has become a target for a few carbon-based lifeforms on ICC. What
>>>I've ascertained is that some (lower-rated) players will play the program over
>>>and over in blitz until they manage a win against it.  Then they play the
>>>same game over and over and grab the rating points.  It works because a win
>>>by the program only nets back a point or two while a loss transfers a large
>>>number of points to the lower rated player.  The program, being only as smart
>>>as its programmer, happily plays the same (losing) game over and over :(
>>>
>>>I'm actually surprised that anyone takes their ICC rating that serious, but
>>>apparently they do.  Of course I could no play them, but what's the fun in
>>>that?  Besides that seems cowardly ;-)
>>>
>>>Learning would get rid of the problem entirely and make the program's book
>>>better as a bonus.  So when I get back from the artic reaches of Minnesota
>>>(it's work-related, don't ask) learning is next up on the docket.  I guess I
>>>should kill the program until I get back home.  I was going to let it play
>>>for the next two weeks continously, but I'm not so sure now (or maybe I
>>>should just let it limbo, "...how low can you go")
>>>
>>>regards,
>>>--tom
>>
>>hi tom,
>>
>>i have seen the same behavior of the low-rated players. in fact, this constitues
>>abuse as far as ICC is concerned - you can report them and they will lose their
>>rating points again :-)
>>but when this happened to me, i never reported anybody. after all, these crazy
>>guys are in fact great testers: they will try and try and try until they find a
>>way to beat up your program. which isn't all that bad, because mostly there is
>>something to learn from that...
>>i solved that type of problem more or less by adding a very broad opening book
>>to my program, now the abusers at least can't shoot at a sitting duck any more.
>>
>>cheers
>>  martin
>
>Hey Martin,
>
>Naw, I don't want to report anyone either.  I'm really not that hung up on
>ratings but did find it interesting.  I've had "implementing learning" on
>my TODO list for a while now, so this will be the final impetus I need to
>get off my duff and finish it.  Like you I also consider them great testers,
>(these humans, in general, are a tricky lot ;-)
>
>Also anyone trying this in the future will be a great acid test for any
>learning algorithm(s) I implement.
>
>regards,
>--tom

The issue here is whether you will try to codify an existing learning algorithm
or try to invent one yourself.  The latter might be called "reinventing the
wheel" but could be worth the effort if you come up with something original and
better.  It is a tradeoff to determine the best use of your available chess
programming time.

Bob D.



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

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