Author: Don Dailey
Date: 20:57:14 12/13/97
Go up one level in this thread
O>Particularly if the test is not published. Test positions tend to end up >cooked and then the set is not valid anymore. :( That's part of the reason bigger sets are better. >As for the 1,000 positions, I think that a test that includes many >positions may be more accurate, but also too time consuming. Yes, I agree, the more problems the better but long sets will tend to not get run. The testing procedure will probably be to run each position for 60 seconds. But building the set will be difficult. The only way this method will work is if several progams are run against the intitial data for long time controls. It's unclear whether there is enough interest and motivation for doing this yet, but I can give it a shot. >A criteria I followed when building my test was to find positions as >unambiguous as possible, with one good answer and only one. Still, >programs hesitate, at times going back and forth. Yes, this is an important quality that a good set must have. I hope the "conset" idea which is designed to produce this kind of problem works out. During the first pass, Cilkchess will agree with the master about half the time I am guessing (rough guess) Then I'll throw out the ones that are completely obvious, or are solved in less than 1 or 2 seconds. Of the remaining problems, many will have more than 1 "best" move available, and Cilkchess will have picked the Master move by accident or coincidence. We will find these positions when other good programs disagree on the move and we will throw these out too. The positions that remain will be only the ones all programs agree upon after long searches. I think almost all of these will be truly best moves that fit your criteria for goodness and uniqueness. But we will see. We can always look at the final results and say that it didn't work out too well, or that it was a success. There are other ways I can separate these positions too but this way gets everone involved and shares the work load. I think this is good. If no one helps out then no harm is done. I'm viewing this as an experiment anyway. The point is to give it a try and see what happens! Maybe something useful will result. -- Don
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.