Author: Anthony Cozzie
Date: 17:47:39 10/24/04
Go up one level in this thread
On October 24, 2004 at 18:33:17, Brian Richardson wrote: >I am again looking at adding TD learning to Tinker. >Thank you to those that have provided comments and suggestions thus far. > >My concern is that the position that the final score is based upon >must be saved. This is awkward to do, it seems. So, I walk down >the PV as far as possible, and then look at that position. > >Unfortunately, the PV does not normally extend far enough, due >to quiesce search, or other search "instabilities". > >The good news is that I found several PV and search bugs, >and things have improved. > >The bad news is that for TD learning, sometimes >the final score does not match the evaluation score >for the walked PV position. > >Then I tried matching the final score with a qsearch score from >the walked PV position. This almost always matches, but not _all_ >of the time. >For example, for Tinker, running >8/8/7k/8/4p1K1/8/5P2/8 b - - Fine16 bm e3 >nothing matches after 12 ply, but then things stabalize and match again >for awhile, and then there are more mismatches, and so on. > >I have tried testing with and without any hashing, pawn hashing, >force stuffing the PV into the hash table after each iteration, >and some other basic things, but there just seem to be a few cases where it does >not match. > >My question is for those that have already added TD learning to their programs, >was this a problem, or perhaps your engines have a "cleaner" PV? > >I could just run with qsearch instead of eval, but of course that would add >quite a bit of time to the learning computation runs. > >Thanks, >Brian You could also give the annealing algorithm a shot - I've had good luck with that, both with Zappa and Crafty. Search the CCC archives. IMO comparison training is the most feasible method of evaluation tuning. anthony
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.