Author: Steven Edwards
Date: 16:32:28 05/07/05
Go up one level in this thread
On May 07, 2005 at 19:13:42, Tor Alexander Lattimore wrote: >On May 07, 2005 at 09:39:10, Steven Edwards wrote: >>Here's WCSAC.0050: >> >>[D] r1bq1rk1/pp3pb1/2p1pnp1/4NnBp/2BPN3/2P2Q2/PP3PPP/R3K2R w KQ - 0 1 >> >>Book: g4 >> >>Toolkit: Rd1 Qe7 g4 a5 gxh5 gxh5 Rg1 (+2.798) >> >>Currently running a probable cook rate of about 8%. > >does this mean you plan on removing the position or simply adding new solutions? >It looks to me like my program might play Rd1 by chance and while it can be set >such that a solution only counts as correct with a high enough score, but that >is dodgy in some positions also I would think. Adding solutions is the way to go; removing positions is a bit too severe, I think. No matter what the soution set is, there is always a chance that a program will match one of the best moves accidentially. A random move player will still score about three percent and there's nothing that can be done about this. But it's not really a problem as long as there are a good number of test positions with answers that strongly correlate with best play. The same situation exists with aptitude and various other standardized psychometric exams. >Perhaps an idea would be to have a new type of 'bm' flag that instead of best >move suggests a move/s that *must* be in the pv or maybe a generous indicator of >a likely score range. Perhaps, but how would this data be established in the first place? My approach is less subjective: run test suites at time limits that are two or more magnitudes larger and on multiple strong programs and use these answers.
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.