Author: Stuart Cracraft
Date: 09:26:46 03/05/99
Go up one level in this thread
On March 05, 1999 at 00:53:22, Jeff Anderson wrote: >A new article in Chess Life by Larry Kaufman called "The Evaluation of Material >Imbalances" is very interesting, and should be read by every computer chess >enthusiast. Using a large database of master games, Kaufman undertook to fine >the real value of the different chess pieces, and various combinations of chess >pieces. It would be interesting to incorporate what he says into the evaluation >of a chess program. One could easily do it with ExChess or Gromit, where the >evaluation is very easily editable. One of them most interesting things that >Kaufman proposes, is that the 2 bishops is worth half a pawn! I would be very >interested to see what would happen if Mr. Hyatt made the bishop worth half a >pawn in Crafty's evaluation. > >Kaufman gives this as the basic table of values for pieces: >pawn = 1 >2 bishop bonus = .5 pawns >B = 3.25 pawns >N = 3.25 pawns >R = 5 pawns >Q = 9.75 pawns > >The article talks about other subjects such as the value of the exchange, 3 or 4 >pawns for a piece, two rooks vs a queen, and pieces vs. the queen. He even >suggests that a rook pawn is only worth .85 of other kinds of pawns. > > >Further research into such subjects with large databases would be very >interesting too. I remember when KK said that one should avoid trading queens >because white's winning probability goes down 1% after doing so, and as I recall >it stirred up quite a controversy. >I would very much like to hear other's comments or thought's on the article. > >Jeff I think a lot of it depends on how much credit is given for piece mobility and piece placement. These can easily corrupt the above material imbalance numbers on any program. For example, one often sees some computer programs trading a knight and a bishop for the pawn at f2/f7 followed by the rook defending that pawn in the castled position. Typical material values would have that 1P+1R < 1N+1B (e.g. 6.00 < 6.50) but there may be extenuating circumstances. The rook may have good mobility along its rank whereas the knight and bishop may have poor mobility. Plus usually a huge bonus is given for pawns in front of a castled king so by taking one of those away via the above exchange, well you see what I mean. However, that kind of capture is usually not a good thing. GNU Chess 5.0's current piece values are (normalized to above): P=1.00, N=3.5, B=3.75, R=5.5, Q=11. This prevented such exchanges as above in its play on FICS. --Stuart
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.