Author: Dan Homan
Date: 07:10:16 07/28/99
Go up one level in this thread
On July 27, 1999 at 19:35:43, KarinsDad wrote: >On July 27, 1999 at 19:15:04, José de Jesús García Ruvalcaba wrote: > >[snip] >> >> Maybe Crafty needs to study «The art of the middlegame» by Keres and Kotov. It >>has an excellent section onf «How to defend difficult positions» by Keres. >>Highly recommended. He does not care if a position is theoretically lost or not, >>the goal is to make the opponent's task of winning as hard as posible. >> Includes in-depth analysis of a critical position of a game between Capablanca >>and Rubinstein (St. Petersburg 1914, I think), a Queen ending in which >>Rubinstein was a pawn up. Capablanca defends tenaciously and draws the game. The >>critical position is one in which all the critics said Rubinstein let the win >>slip away. Keres explains Capablanca's defensive *plan* after the move suggested >>by the critics, and then backs it with tons of variations. >> I think the study of these analysis alone make the book worth having. I wonder >>how these concepts can be incorporated to chess engines, as yours seems >>completely different it might be more natural for you to include them than for >>the other chess programmers. >>José. >> > >Interesting point. > >It is important to note that a computer playing a computer cannot swindle the >opponent which effectively is what happens when a person is successful in >"tenaciously defending" against another person. > >Knowing that a program is playing a person as opposed to another computer can be >a good reason to increase the "contempt" factor (where contempt is not exactly >what it is currently, but rather one where the program does not always expect >the opponent to come up with the best possible move in a complex position, so it >can make slightly inferior moves which complicate the position or startle the >opponent). > >My program will (eventually) have multiple candidate moves and pick the move >which appears to follow the plan of the game as opposed to the one with the best >score (although the scores have to be fairly close, 1/10th pawn or so, or it >doesn't make sense). This means different things in closed middlegames, open >endgames, and in a variety of other types of positions, all of which have to be >painstakingly analyzed and recognized ahead of time. Very interesting idea! Having multiple candidate moves sounds computationally expensive though... are you going to search each move at the root separately and resolve a score for each? Another idea for following a plan would be to pre-process a bunch of information and fill piece-square statements or values to conditional statements which will be used in the eval while searching that position. As you point out, deciding how to make a program follow a pre-determined plan (or even to find the plan to follow) is a very difficult task. Sounds interesting, though. - Dan > >Thanks for pointing out this game as it may serve as a catalyst for one or more >plans for my program (and of course, you gotta love the endgame playing style of >Capablanca). > >KarinsDad :)
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.