Author: Robert Hyatt
Date: 11:02:22 06/29/99
Go up one level in this thread
On June 29, 1999 at 11:43:13, Albert Silver wrote: >On June 29, 1999 at 09:44:23, John R. Menke, Sr. wrote: > >>----------------------------------------------------------------------------- >> >>In March 1999 Chess Life magazine, IM Larry Kaufman presented his research into >>the relative material values of the chess pieces based upon statistical research >>using a database of nearly 300,000 games where both players had FIDE ratings of >>at least 2300. His summarized values were as follows: >>Pawn = 1 >>Bishop Pair = +1/2 >>Bishop = 3 1/4 >>Knight = 3 1/4 >>Rook = 5 >>Queen = 9 3/4 >>He says these values agree with the statistics within about 1/8 pawn accuracy in >>nearly every case tested. He also suggested a possible slight bias in favor of >>the Queen, which could be valued at 9 1/2 instead. I'm curious if these values >>have been tested with chess playing computer software vs different values? >> >>With Chessmaster 6000 I have come up with the following approximation which I'm >>testing, and have dubbed it the "Kaufman-Menke" values: >>Pawn = 1 >>Bishop = 3.3 >>Knight = 3.2 >>Rook = 5 >>Queen = 9.7 >> >>In comparison I believe that Terry Ripple is using the following values (dubbed >>"CM6555") for his tests with CM6000: >>Pawn = 1 >>Bishop = 3.5 >>Knight = 3.3 >>Rook = 5.5 >>Queen = 10 >> >>And I understand that Shep's values are as follows: >>Pawn = 1.1 >>Bishop = 3.5 >>Knight = 3.3 >>Rook = 5.5 >>Queen = 10 >> >>In a related note, I recently encountered the following interesting position >>which I believe might be a useful benchmark position in some of these tests >>regarding the relative values of Bishop and Knight. Slight changes in their >>material values will determine the subsequent course of the game. In one case >>the Bishop will be exchanged for Knight, or it will be retreated because it is >>slightly more valued. The position arises after the following moves: >> >>1 d4 Nf6, 2 Nf3 d5, 3 e3 e6, 4 Bd3 c5, 5 c3 Nc6, 6 Nbd2 Bd6, 7 0-0 0-0, 8 Re1 >>e5, 9 dxe5 Nxe5, 10 Nxe5 Bxe5, 11 Nf3 Bg4, 12 Be2 ... >> >>The question is: Should black play 12...Bxf3 or retreat the bishop to d6 or c7? >> A slight change in the relative material values of the Bishop and Knight causes >>Chessmaster 6000 to make different decisions here. >> >>All comments welcome... >> >>John Menke >> >>----------------------------------------------------------------------------- > >I'm testing this out with Nimzo2000 running through Winboard. Nimzo2000's >default values are: > >Pawn = 105 >Knight = 425 >Bishop = 435 >Rook = 685 >Queen = 1240 > >BTW, another fascinating article on material values is Soltis's article >susbsequently republished in his book "Karl Marx Plays Chess". It's very >interesting and quite funny sometimes. > > Albert Silver Those are actually a kludgy way of stopping two minors for rook + pawn, or minor for 3 pawns, and so forth. But it causes other problems. I think it is better to simply 'evaluate' that a piece for 3 pawns is good for the side with the extra piece, or a rook+pawn for two minors is better for the side with the two minors, etc... I tried all sorts of oddball scores, and finally wrote what I thought was right in the first place. IE R+P+P == 2 minors. And Larry found this in his article as well. Also, P+P+P+P == minor, which he also found and reported... Been using this 'bad trade' code for well over a year and haven't had problems since I added it. Before that I was constantly trinking around with the piece values to stop bad trades that way. And minor positional scores would make it take/avoid such trades in odd places. No more...
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.