Author: Robert Hyatt
Date: 07:55:23 01/20/98
Go up one level in this thread
On January 20, 1998 at 02:53:37, Amir Ban wrote: >On January 20, 1998 at 00:54:39, Don Dailey wrote: > >>On January 19, 1998 at 22:31:53, Bruce Moreland wrote: >> >>>I think that everyone who's involved in this wants something different. >>> >>>I want a tactical suite that is challenging and doesn't take fifteen >>>hours to run. >>> >>>I think you want something that is more correct than this. >>> >>>bruce >> >>Ok everyone, let's reach a consensous. Fill out this form: >> > >Ok, but please take netscape editor realities into account when >designing forms :) > > >>YES NO >> > > x > >>---- ---- 1. Should the program have a single clearly best move? >> > > x Isn't this just the converse of 1 ? Should be all moves >of equal merit, not so much in term of eval as in terms >of deciding the game result in equally fast & safe ways. > >>---- ---- 2. Can it have more than 1 clearly better than others? >> > > x Yes. This suite doesn't have mate in 2's like WAC, and >one > man's two-plies search is another man's 5 minutes think. > >>---- ---- 3. Should we KEEP the easy problems? >> > > x The only moves that cannot be found by accident are >sacs, and I don't want a sac-only suite. > >>---- ---- 4. Is it ok if programs tend to solve it by accident? >> >> >>Here's my answers: >> >>1. YES but don't care that much >>2. NO but don't care that much >>3. YES but don't care that much >>4. NO but I can cull them later >> >> >>I believe point 4 is the only one anyone feels strongly about but I'm >>not sure. If this is the case let's do the set without worrying about >>this point. Later we can decide without hurting a thing. I can cull >>these out later and if anyone else wants this they can either help me >>or wait for me to do it. >> >>But we've done good work so far and identified a bunch of bad problems >>so let's get moving! We should post a list of the whole set minus >>the bad problems once every day or two. Certainly we can start with >>the incorrect ones which we tend to agree on as Amir notes. >> >>P.S. Who is going to post the set (minus the cooks?) I can do this >> if no one else is going to. Ed volunteered to use his site >> as a repository and I concur. >> > >Please go ahead. I assume you have noted all posted data. I didn't see >any conflicting opinions yet, but nobody has yet confirmed Dark >Thought's Rf4 cook for no. 29. > >Amir Note that this is a sac-only suite for the most part. It does not claim to be a "positional test" at all. My only interest is to try to come up with something that is not unreasonably long, which can be used to compare programs and how they handle tactical positions. IE Bruce and I compared notes on WAC for a long time until it became obvious that WAC wasn't useful any more, since we both started getting over 290 at 1 minute or less. I don't like a 5 second suite as that is *really* getting into the noisy part of my search. I don't like 15 minute suites either as that takes too long. I'd like 300 positions at 60 seconds per position where I start off at maybe 50 right or so. Then I have something to measure future tactical improvements with. We do need a positional test suite. However, *every* position I have looked at in ECM is tactical, based on a capture or a check or a tactical threat. That's only one part of a chess program. But it seems to be the only part that ECM is going to measure..
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.