Author: Roger D Davis
Date: 23:55:24 06/28/99
Go up one level in this thread
For me. the unclear versus equality thing is only one small factor in the question: "Can the backsolving of the same opening book be replicated by different programs?" For me, that is the question. Now the answer. Backsolving depends on migrating leaf node evaluations backward through the tree. Thing is, when I analyze even one game with different programs like Fritz 5.32, Hiarcs 7.32, Junior 5.0, Chessmaster 6000, and Rebel 8, I often get wild variances in the evaluations between programs, especially at crucial points in the game. And it's exactly those numbers that the program is using to backsolve with. I really can't imagine taking that variability and multiplying it by all those programs, and then expecting my opening book to be backsolved in the same way. Moreover, Bob and later the Deep Thought guys showed that programs continue to discover good moves with some frequency no matter how long they're allowed to think (that's an overgeneralization, I just can't remember the specific number of plies these experiments used). So, if you have a book of some size, I would imagine that you could obtain substantial variability in the way its backsolved even within a single program if you use, say, 2 minutes of time per move versus 4 minutes of time per move. That doesn't sound like much time, but if you database is huge, it quickly adds up. So...IMHO, backsolving is best used with small opening books where each move has been analyzed to death by a program that plays in the same style you want to learn. Since you can expect variability in evaluations between programs, this last part is important, because you are teaching yourself to play like the program that did the evaluating. If it doesn't suit your style, you'll follow lines that don't suit your style, and be trashed coming out of book. Roger On June 28, 1999 at 18:46:49, Chuck wrote: >On June 27, 1999 at 20:33:24, Jerry Creed wrote: > >>After re-reading Komputer Korners explanation of violations >>of the opposite priority rule leading to eradication of =/ and /= Informant >>signs and the non-capturing all of the numerics phenomenon when importing >>EPD files, backsolving seemed even less usable than ever, unless I wanted to >>not use it and do all evaluations manually. However, this dilemma is >>clearly neutralized when re-reading Mike Leahy's Backsolving de-mystified. >>So, can backsolving be trusted >>if left on to Always Solve, should favor unclear over equality be checked or >>unchecked and would you allow backsolving to work on Informant symbols or >>just plain numerical, as from Zarkov? >> >> http://www.icdchess.com/wccr/index.html to find Komputer Korners' report : >>Bookup 1.5.2: A Continuing Review by Komputer Korner on Wed May 12 11:53:24 >>1999 >>http://www.bookup.com/ Home of Bookup >> >>jcreed@snip.net > >I think Bookup does what it says but sometimes these results can be surprising. >I wouldn't trust the evaulations without looking at the lines which follow. As >far how a position should be valued, I think this is in large part a matter of >taste, personally, I don't see much difference between "unclear" and "equal". It >depends on who's doing the classifying.
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.