Author: Robert Hyatt
Date: 10:34:52 05/19/00
Go up one level in this thread
On May 19, 2000 at 12:11:51, Dann Corbit wrote: >On May 19, 2000 at 11:24:51, Robert Hyatt wrote: >[snip] >>I think a deep book is suicide. Every move made by a human has a probability of >>error (being wrong) that is > 0. The deeper the book, the more moves there are. >>And probability is summed across all the moves in that game. Yes you will stay >>in book for a very long time. But every time you come out, you realize you are >>lost because the opponent varies when he finds something that was missed in the >>game used in the book. >> >>I have done this before. I gave up on deep books as a result, unless you have >>the machine time to invest in searching _every_ move that goes into the book, >>_very_ deeply... And even then you will overlook even deeper tactics that a >>human might spot. > >I think it obviously requires a new paradigm. For instance, your analysis could >increase for each move forward so that by 30 moves you are fully analyzing >everything. That way, you don't just fall out of the book. You come out >gradually. > >I think that you have another possible thing to try also. > >Suppose that you are at a branch where there are four different courses taken by >humans. Start 4 threads analyzing each of those possibilities and ignore the >others [if, for instance, there are 35 choices we are only going to look at the >4]. The objective is not to choose the best possible computer move but rather >to make a program that plays human-like chess, but very strongly and avoiding as >many mistakes as possible. This is what "book random 0" is all about. :) I use this in important games. > >There are always alternatives to test new information, or different ways to >think about applying it. > >You are right about the big book making the program significantly weaker. >That's the book I choose when I want to play against it and I want it to play >like a human. It still does not play stupid chess. It picks the choice that >wins more often, as you know. Yes... until you get way into the book, where there are few games. If you have two choices, one that won and one that lost, playing the one that won is a coin- toss. It could have been won due to blunder by the opponent, while the one that lost could have been lost by a blunder by the player.
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.