Author: Victor Zakharov
Date: 03:54:24 06/24/03
Go up one level in this thread
On June 23, 2003 at 13:17:57, Helmut Conrady wrote: >On June 23, 2003 at 10:50:21, Victor Zakharov wrote: > >>>>>Can someone help me with links leading to the "computer analysis project"? >>>>>Thanks. >>>>> >>>>>Helmut >>>> >>>>There is no a lot of people who are ready to provide their computers at nights >>>>for CAP project. Technically CAP tools are done and there are no problems to >>>>include into the project thousand of computers. Currently 20-30 computers >>>>permanently work for the project and there is about 13 millions positions >>>>analyzed with different engines like Crafty, Tiger, Ruffian and Shredder. >>>>CAP database is supplied with Chess Assistant as one of CA trees. >>> >>>Yes, I know. I own CA 7.1. >>> >>>What about the time control in this project. Always the same? Infinite analysis >>>mode? >> >>The main problem of the current data is that they are unbalanced. Some positions >>were analyzed a few years ago with Crafty for 1 minite on Celeron-400 computer. >>Another - for 3 minutes with Tiger and Shredder on 2GHz computer. Different >>engines can evaluate the same moves strongly different. Some lines are deep and >>some lines are short. So any who uses CAP should look at it critically. >> >>CA team implemented a set of procedures that will make CAP well balanced. > >Can you please explain this a little bit more? There was a few stages in CAP development. First CAP database contained evaluations of all positions that met at least in 50 games. There was no a big benefit from this. Today CAP contains analysis of all positions that met at least in two games. Making tree balanced include two aspects: I) Searching for actual problems. II) Improving analysis. I. If some evaluation correlates with reasonable evaluation there is no need to analyze it more. So the first step is searching the places where CAP evaluations contradict with 1) Master evaluations (i.e. Eval tree in CA7). 2) Statistics (as percent of success, so the number of games). 3) Bad engine result in some opening. II. Additional analysis includes the following procedures. 1) Analysis with latest engine and longer time. 2) Analysis of final positions in the subtree. 3) Prolonging analysis of the main line on specified number of moves 4) Analysis of actual games after novelty on specified number of moves 5) Manual analysis. There is set of functions in CA7 that aimed for the purpose of improving evaluations. For example - autoplay for some number of moves, - searching for alternative move candidates in some range of evaluations and line autoplay with these moves, - retroanalysis of variations - automatch of two engines for specified number of moves. This one is used when two engines evaluate position strongly different. So every engine play for the side that it evaluates better. - Deepen analysis along best lines. This operation selects subtree with some range of evaluations and send all positions including final ones to analysis. I think I should finish on this. Otherwise it will too many lines here. Victor
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.