Author: Patrik Lindqvist
Date: 22:33:47 05/03/98
Go up one level in this thread
On May 03, 1998 at 13:13:28, Don Dailey wrote: >On May 03, 1998 at 03:22:23, Patrik Lindqvist wrote: > >>Im using a program there you can change the values for the chess pieces >>from 0-2000. In standard settings which is for normal play the settings >>are Pawn=100, Knight=420, Bishop=440, Rook=685 and Queen=1300. Since i >>mostly play blitz game i was wondering if there is anyone who has any >>pointers or tip on what chess pieces values that are suited best for >>that kind of games. > >Without understanding this program you cannot easily tell what good >values are without much experimentation. His piece values seem really >high to me but might not be if he uses large pawn evaluation bonus's >or converserly tends to use heavy penalties for the non-pawns. > >Some experimentation can tell you a lot. But here is my advice if >you want to play with values: > >The most productive way to make an attempt to improve the program, >given that you only have these 6 "knobs to turn", is to experiment >with the relative difference between the knight and bishop. > >Almost every game involves this trade. Is extremely common >and often one decision is better than the other. Being off just >a few points in either direction might make the program significantly >weaker. Or if it currently off, you might improve it significantly >by a small tweak in the value. > >In my chess program tests have shown that there is a small window >of score differences I must honor to squeeze the most strength out of >the program when it comes to the difference between bishop and knight. >For my program it seems to be that bishops and knights score the same, >(but I have a significant bishop pair bonus.) Also a knight pair >penalty is useful, because if both sides have a bishop and knight >then trading one off for the other is no big deal (depending of course >on other factors.) The worst case is having 2 knights vs 2 bishops. >A master once told me this was pretty much like being down a pawn! > >So all things considered, it turns out that bishops on the average >are worth more than knights but if you can isolate the important >factors and treat them separately, this is better. My program considers >piece cooperation terms as material terms not positional. My latest >thinking is that the "raw" bishop actually is very slightly more >valuable than a kinght, so my program will slightly favor 1 bishop >vs 1 knight, all other things being equal (whatever that means!) > >But you must realize I am speaking only in generalities. There are >many terms in the program that will modify the values of these >pieces and change everything! An example is the presense of pawns, >especially center pawns which tend to increase the knight and >decrease the bishop. > >So the "correct" values for the pieces would have to be determined >individually for each program. There is no simple way someone could >provide you with a new table of values to plug in and expect it to work >well unless that person knew the program well. > >By the way, which program is this? If it's a strong program there >is a good chance the default values will be very hard to improve upon. > >- Don The program is Nimzo98. Chrilly Donninger, who wrote the program, told me that the default values are for normal play but he thought that those values are not optimal. And for blitz games he thought there is definitly some better values for the program.
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.