Author: Dann Corbit
Date: 15:23:11 08/26/02
Go up one level in this thread
On August 26, 2002 at 16:19:34, Uri Blass wrote: >On August 26, 2002 at 16:16:35, Matthias Gemuh wrote: > >> >>Hi Uri, >> >>>I think that you are clearly wrong here. >>>It is easy to get the opponent out of book in a few moves >> >>I would say that it is possible, not easy. >>And the opponent can later transpose back into book, if book lines are long >>enough. >> >> >>> >>>The engine need to play later. >>> >> >>The problem is of course, that they sometimes have no chance to play. >>Junior 6 once left book at 3 moves to being checkmated by Fritz 6. >>Cases of checkmate while still in book have been reported. >> >>Regards, >>Matthias. > >This can happen only if you have big books that were not checked manually. I think that cases of checkmate can be checked more reliably automatically than manually. Run chest over all the positions in the book at depth = 5. (This is very fast) Repeat for whatever level of safety you desire. I think also it is common sense to analyze every exit point from your book. If you have a million positions in your book, you might have only 50,000 exit points. So analyze them on two seconds time control (would take one weekend). Then, any positions that are more than 1/2 pawn from zero, analyze at ten seconds. Continue this sort of process until you are satisfied with the quality of the answers. Now, having performed this automated analysis, I think manual analysis could be performed. In order to do this, have the chess engine spit out the "supposedly" optimal lines and check them against NCO, MCO, ECO, etc. Since you only have 20 of them from the starting position, it should not take too much time. After each correction, re-run it and re-check it. I think a big benefit can also be had by a retro-computation for all of the supposedly optimal lines. For each of the 20 exit points, analyze for two hours on the fastest machine you can get your hands on. Takes one weekend. Then, if any of the positions reports something strange, do a deeper analysis. This should give a book with enough confidence to use it in a contest. Not as good as a professional book (of course) but much better than something compiled under pure automation without rechecking afterwards. Tony Werten has a very good idea about books. I hope he publishes his findings.
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.