Author: Bas Hamstra
Date: 14:17:06 01/06/01
Go up one level in this thread
On January 06, 2001 at 14:34:09, James Swafford wrote: >On January 06, 2001 at 12:11:32, Bas Hamstra wrote: > >>Hi James! >> >>On January 06, 2001 at 11:48:44, James Swafford wrote: >> >> >>I started with all weights zero. To see it did something reasonable, I let it >>play a pool of 100 slightly randomized hand-tuned evals. You don't need 100 >>programs, just 100 weight sets and one program. It then learns to score 50% in >>200 or 300 games. I plotted the weights real time in a graph to see how they >>developed. This was my first test. Second step was to put it in my console based >>"real" program and it is playing Crafty right now. > >Seems to me it would help things along if you started with reasonable >values. I think Tridgell / Baxter started with all weights = 0, too, >with KnightCap, and remarked the same thing. >>>2. What do you mean by "wrong trend?" I suppose you mean a term >>>is "drifting" the wrong way... becoming more negative when it should >>>be going more positive? >> >>Yep. Say 90% of the weights tend to show reasonable values. But a few don't at >>all. It might be that it needs more games, though. I am not sure if this is the >>fastest way of automatic parameter tuning. Maybe some kind of "weight fitting" >>on a large set of positions is more efficient, but how? > >What would happen if you "moved" those weights to where you think they belong? That would be cheating :) I want to see it work straight away and then see the score percentage go up. >>>3. How are you training your evaluator? With a wide variety of >>>opponents, or by playing the same programs over and over, or ??? >>>How many games have you played? >> >>Right now I am playing Crafty for a couple of hundreds of 1 0 games. > >Hmmm.... so you're training your evaluator to play the best it can >against Crafty at 1 0. Why not ICC (or FICS) against a wide variety >of opponents? I'd prefer slightly slower time controls, too, although >I know with that many games time gets to be a problem... Yes, I want to have a reliable method first. I do not feel I have reached that point. One the method works for 1 0 it will work for longer tc's too. The nice thing is you can MEASURE if it works. If it works, the score must go up. >>>4. Does your engine compete on ICC? >> >>A couple of times. But mostly FICS. The TD version has not played there yet. By >>the way: I compared what it learns from a) wins b) losses c) draws. In my >>opinion a) and c) did not do well. So now I only learn from losses. Never change >>a winning team. > >Yes, that makes sense. Thanks for the info! Maybe we can share notes >in a few months... One additional advantage is that when learning from losses only, you don't have to deal with blunders that distort learning. The computer makes no (tactical) blunders. And a blundering opponent will not win against a computer. This (I suspect) increases the probality that IF it learns something, it will be something useful. Ciao! Bas.
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.