Author: Eric Oldre
Date: 15:19:57 01/06/05
Go up one level in this thread
On January 06, 2005 at 14:49:06, Dann Corbit wrote: >On January 06, 2005 at 13:20:27, Eric Oldre wrote: > >>I have added the ability to add opening books to my latest version of Latista. >>I'll release it once I create a decent opening and add a couple other smaller >>features. >> >>So far I created a book my parsing the gm2600.pgn file and saving all positions >>which occur at least 5 times. >> >>This is certainly much better than no book at all but leaves much much room for >>improvment I'm sure. >> >>What general advice do you book makers have for me? > >I think that correspondence games between top players are best. >Next comes 2600+ games from players after the year 1960. >Then comes SSDF games on the Athlon 1200, where both programs are top 15. > >You should also weed out the very short games. > >Finally, filter for only moves played successfully at least 3 times. > >Then you will have a decent competition book, with less than 1% blunders in it. Dann, Thanks for the tips!!!! Very useful. My short term goal is simply to create something that will get my engine to play a variety of openings and not make any huge blunders right off the bat. while saving time on it's clock. So for Latista 1.3 I'll probably need to stop after the above. Your suggestions below are really interesting to me. I have noticed that sometimes latista comes out of book thinking it's well ahead, and sometimes well behind. I had wondered if coming out not liking it's position could hurt it. even if the position really is ok, if it's not the "style" that my engine normally likes than it might do things that are counter productive in it's first few moves out of the book. I haven't had enough experience yet though to determine if that's really the case. but it seems like it would make since. And doing what you described below would certainly help the situation if that is the case. Getting things set up to where I could run something like that would be quite a job. I'm saving this thread though and hope to get to it eventually. Thanks, Eric > >If you have the time, analyze each exit point from the book at as long of a time >control as you can stand with your program. > >Suppose (for instance) that you have 10,000 exit points and one week to prepare. >Assuming you have a machine to dedicate to the search, in that case, you could >analyze each exit point at: >(86400 seconds / day * 6 days)/10000 exit positions = 52 seconds. >If you have two machines you can use, then you can get 104 seconds, (etc.) > >Now, I left one day for the 100 positions which will turn up with bad scores. >Take these and analyze them at 86400/100 = 864 seconds (double it if you have >tow machines, etc). >If they are still bad, mark them as "don't play" since your engine does not >understand them or they really are bad choices. But you still have a problem. >The position that lead to that move is still in your book. So you will need to >mark all positions that lead to that move and which do not also have a choice to >lead to a good one as bad. > >This (of course) will still leave a few hundred bad positions in the book. The >only way to weed those out is with book learning and/or human analysis. Either >one of those will take a long, long time. That's why the big boys hire >professional book builders. > >Another you can do (if you have time) is to minimax your book, to tag optimal >trails.
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.