Author: Peter Fendrich
Date: 14:03:40 08/15/01
Go up one level in this thread
On August 15, 2001 at 12:59:21, Bruce Moreland wrote: >On August 15, 2001 at 06:52:18, Peter Kappler wrote: > >>On August 15, 2001 at 05:03:20, Peter Fendrich wrote: >> >>>Has anyone tried to clean up the ecm test suite? >>>//Peter >> >> >>There was a big effort here (or was it rgcc?) a few years ago. I don't know if >>anybody went through the effort of verifying all the fixes and publishing a >>"clean" version. I remember that many of the fixes/alternatives were very deep >>variations and took a lot of CPU time to verify. >> >>Bruce Moreland has mentioned that he likes ECM -- he might have been part of >>this effort. Probably Bob was in on the action, too. I can't recall the other >>participants. >> >>It would be neat to try cleaning it up again. Machines are much faster now. :) >> >>-Peter > >The problem with that effort is that we didn't define what we were trying to do. > Were we trying to cull bad positions? Were we trying to assure singular >correctness of positions? Were we trying to assure that the key was the only >move that won to some degree or another? Were we trying to find hard positions >that were correct? > >There are lots of types of suites that would be of interest to people. A suite >comprised of hard positions that can be solved with present techniques in >between thirty seconds and an hour would be very useful, and would probably be >useful for several more years. > >The problem with ECM is that it takes a long time to run to any appreciable >depth, some of the problems are wrong, some have cooks, some end up being found >for "positional" reasons, and many are way too easy. > >If I take the ECM positions that Ferret doesn't find in 20 seconds and make a >suite out of them, I end up with a suite that almost by definition everyone else >will do better on. Yes, but if "everyone" did the same we could match the result and find the wrong ones together with the really hard ones. By continouing in the same way with this suite with longer time frames we would probably have a good idea of which are the wrong positions. This could be step 1 before any grouping takes place. > >I think it might make more sense to grade the positions somehow, and break the >suite up into pieces that are easy or hard, get rid of the obvious mistakes, and >get rid of the ones what cause positional arguments. > >I don't have time to do this now since my plane leaves tomorrow. It's a long flight and you must have something to do, right? :-) //Peter
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.