Author: Omid David Tabibi
Date: 11:57:12 12/13/02
Go up one level in this thread
On December 13, 2002 at 14:32:09, John Merlino wrote: >On December 13, 2002 at 14:10:02, Omid David Tabibi wrote: > >>On December 13, 2002 at 12:25:55, John Merlino wrote: >> >>>On December 13, 2002 at 11:11:07, Omid David Tabibi wrote: >>> >>>>On December 13, 2002 at 10:44:47, Vincent Diepeveen wrote: >>>> >>>>>On December 12, 2002 at 15:56:43, Omid David Tabibi wrote: >>>>> >>>>>[snip] >>>>>>Finding 13.Bb5 should be very easy for every computer. This opening was first >>>>>>introduced in a match between national teams of USSR and Argentina. The >>>>>>Argentinians played this opening on several boards (not expecting 13.Bb5), all >>>>>>the Russian Grandmasters found 13.Bb5 and went on to crush black. >>>>> >>>>>where is the output of genesis? >>>> >>>>White(13): f1b5 >>>>[book] >>>> >>>>;-) >>>> >>>>>Vincent >>> >>>This is the same with CM9000, and 13...Rh7 is also in the book (but no further >>>moves). So, in a sense, the engine doesn't even NEED to find that move. I was >>>using the analysis feature in my previous posts about this position, and this >>>feature does not access the opening book. >>> >>>FYI, the CM8000 book plays 8...Qc7 in this line, and does not have any response >>>to 8...h6. >>> >>>So maybe I DID make (or, rather, GENERATE) some improvements to the CM9000 book, >>>eh? >>> >> >>Now the CM's opening book consists of more data, but is this extra data useful >>(if not dangerous)? It depends. Is CM aware of the drawish nature of the >>Gothenburg variation? For example, if CM is playing for a win, will it "know" to >>avoid the Gothenburg variation and instead head for the classical 8...Qc7 ? >> >>P.S. If black is playing for a win, he'd better avoid playing 7...Be7 in this >>variation, and instead, play 7...Qb6 heading towards the poisoned pawn >>variation. > >Here is what is in the CM9000 default opening book at the moves you have >mentioned: > >After 7.f4 >-- 7...Qb6, 1/4 probability >-- 7...Be7, 1/4 >-- 7...Nbd7, 1/6 >-- 7...Qc7, 1/6 >-- 7...b5, 1/6 >Also, the moves 7...Bd7, 7...Nc6 and 7...h6 are in the book but have zero >probability. > >After 8.Qf3 >-- 8...Qc7, 3/4 probability >-- 8...h6, 1/4 probability >Also, the move 8...Nbd7 is in the book but has zero probability. In addition to the probability, a gross evaluation is also essential for optimal tournament play. For example, if it is playing for a win, it should eliminate all drawish lines (i.e., set their probability to 0) from the list of choices. BTW, under perfect play by both sides, 13.Bb5 leads into a draw, that is why 13.Bc4 has become more popular in the last years. > >The default CM book has always been designed to be eclectic, with the ability to >play both popular lines as well as winning lines (which, in most cases, overlap, >obviously). With CM9000, we used a completely new set of data to create the >book, but we still focused on popularity as well as success to calculate the >probabilities of the moves. No fine-tuning was done to this book after it was >created, however.... > >The import PGN feature of the opening book editor in CM9000 has the ability to >use different parameters to calculate the probablity that a move will be >selected. I will take the liberty of creating a test book using the same data >that was used to create the default book, except using winning percentage as the >main parameter, rather than frequency. This will take several hours, after which >I will report whether there is any significant difference in these two >positions. > ><I guess I have a lot of time on my hands these days....> > >jm
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.