Author: Mathieu Pagé
Date: 06:43:45 04/08/04
Go up one level in this thread
On April 08, 2004 at 06:04:13, Uri Blass wrote: >On April 08, 2004 at 05:59:35, Uri Blass wrote: > >>On April 08, 2004 at 05:27:43, Tord Romstad wrote: >> >>>On April 08, 2004 at 04:51:01, Albert Bertilsson wrote: >>> >>>>Hi! >>>> >>>>I'm about to start coding on my engine again and have thought about improving >>>>the endgame play. I don't like using egtb so I'm thinking about using special >>>>evaluation cases when there are few pieces left on the board. Is this a good >>>>idea or is special evaluation cases a dead end? >>> >>>I think it is a good idea, but it is much more difficult than it first seems. >>>The problem is >>>that it is not enough to write evaluation code which enables your engine to play >>>some >>>specific class of endgames near-perfectly. You also have to adjust the scores >>>in such >>>a way that they are reasonably consistent with the evaluation of similar >>>endgames with >>>a piece or two more. When you add special cases for many different classes of >>>endgames, >>>keeping all the scores consistent quickly becomes a nightmare. >>> >>>Tord >> >>What about trying to translate the scores to expected result? >> >>I have scores in pawns but I think that it is a bad idea and it is better to try >>to have a score for every position by probabilities(win,draw,loss). >> >>having score by pawns is probably good if you want to have fast something that >>plays but I believe that trying to evaluate expected result for every position >>may be better(it is good also for pruning and extensions because you want to >>prune lines when you are certain about the result and extend lines when you are >>uncertain about the result). >> >>Did somebody try to evaluate expected result and not pawns for every position? >> >>Uri > >I can add that in order to be able to win position with 100% probability to win >you may use a second score in case of 100% probability to win(otherwise you may >fail to win positions Like KR vs K without tablebases). > >Uri Hi Uri, I do not do it, but I already tought about it and it seem to me that it can be an interesting way of doing evaluation. The end games parameters(the weights) of such an evaluation fonction could be determine doing data mining on some thousands of games or on egtb. I will eventualy try this idea, but at the rate my engine actualy progress it could be long, before i found the time to do it. In the meantime I would be interested by any documentation or result of experiments about this. If you have any link to such informations it would be fine if you could let us know about them. Mathieu P.
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.