Author: Anthony Cozzie
Date: 13:55:18 12/31/03
Go up one level in this thread
On December 31, 2003 at 13:15:50, Tom Likens wrote: >On December 31, 2003 at 10:22:31, Anthony Cozzie wrote: > >>On December 31, 2003 at 09:30:03, Tom Likens wrote: >> >>>On December 31, 2003 at 07:54:37, Anthony Cozzie wrote: >>> >>>>On December 31, 2003 at 07:53:49, Anthony Cozzie wrote: >>>> >>>>>On December 30, 2003 at 22:04:02, Toni wrote: >>>>> >>>>>>I'm thinking about the values of pieces for a chess program. I have investigated >>>>>>some engines and the values they give to pieces vary. Some examples are: >>>>>> >>>>>>ENGINE knight bishop rook queen >>>>>>Faile 1.4.4 3.1 3.25 5 9 >>>>>>Amy-0.8.4 3.5 3.5 5.5 11 >>>>>>Crafty-19.4, 3 3 5 9 >>>>>>Beowulf-2.2 3.2 3.25 5 9.3 (defult personality) >>>>>> >>>>>>I have to give values for my program and, as I'm not a strong chessplayer I >>>>>>would like to know your opinion. Are these differences important? What values >>>>>>should be assigned for the highest strength?. The same could be applied to score >>>>>>bonuses, but the list is too large. :-) >>>>>> >>>>>>Thanks >>>>>> >>>>>>Toni >>>>> >>>>>Zappa: Knight/Bishop: 3.25 Rook: 5 Queen: 9.75 >>>>> >>>>>anthony >>>> >>>> >>>>P.S. Do a search for Larry Kaufmann's article. >>> >>>These are the similar to the values used by Djinn (a pawn is 1). I let the >>>position dictate >>>wheter the bishop or knight is stronger. The only difference is I give the >>>queen a value >>>of 9.50. I set up an Excel spreadsheet comparing the values of various piece >>>and pawn >>>trades and set the values based on what I thought were good/bad trades. For >>>example, >>>R+4 pawns should be less than the value of a queen, R+P should be less than the >>>value >>>of two minor pieces, 3 minors > queen etc. It was interesting to see >>>Kaufmann's values >>>come out so close. >>> >>>regards, >>>--tom >> >>I really think the queen is almost as strong as the two rooks or 3 minors [very >>close]. Kaufmann does suggest 9.5 as an OK value for the queen, though. Crafty >>implements some bad trade code where the two minors are given a ~3/4 bonus vs >>the rook. Zappa does not, and I've seen it win multiple games when it picks the >>rook+outside passer against two minors. >> >>Another Kaufman idea that Zappa implements is scaling the values of the pieces >>wrt # pawns on the board. With 8 pawns on the board, Zappa views a Knight as >>3.4 or so and a rook as 4.7 - so it is ready to sac the exchange for a pawn and >>some mild positional compensation. OTOH, with 3-4 pawns on the board, Zappa >>views a Knight as 3 and a rook as over 5. >> >>anthony > >I've considered this but haven't done any experiments yet. It seems >to make a lot of sense though. Maybe I'll give this a go in my next >round of experiments. Have you tried any self-play matches using >static vs. dynamic values in Zappa? > >--tom I've never really done selfplay experiments. Zappa is not strong enough to the point where massive tuning is needed. In a month or so I'll probably start addressing the tuning of the evaluation in more detail . . . anthony
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.