Author: blass uri
Date: 15:08:38 09/08/99
Go up one level in this thread
On September 08, 1999 at 12:48:03, Dann Corbit wrote: >On September 08, 1999 at 06:31:41, blass uri wrote: > >>I suggest to analyze comp-comp games from wccc(one hour per move)(it is possible >>to do it by Fritz engines). >> >>the tactical test suite can be to avoid all the blunders that the engines found. >> >>analyzing one game is some days for one computer and we have 106 games from >>wccc(I include the last draw between shredder and ferret). >> >>We need more than a year of computer time to generate this test. >One hour per move is mostly a waste unless there is something special about the >move. Since the payoff takes exponential effort to get better, you will be >fortunate to get one ply more than ten minutes of effort (which is another >reason why I chose 12 minutes). > >It would take C.A.P. just a short time to analyze the games, especially since a >large fraction of the data will be already analyzed. But there is nothing >particularly compelling about these games is there? If we had human analysis to >add on top of it, I think it would be very worthwhile. > >The less than optimal moves most of the programs made I think were largely bugs >in the openings database files some of them used. > >If you want a tactical suite, you can get about 8000 test positions that have >been carefully analyzed from my ftp site under /public_cap_data I want tactical suite from practical games and not from positions that were propsed by humans because I think that avoiding tactical mistakes from practical games is more relevant to evaluate the tactical ability of chess programs relative to positions that were proposed by humans. I believe that humans like sacrifices so the relevant moves is a sacrifice in most of the cases when practically there is often a tactical idea that is not a sacrifice. Uri
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.