Author: Peter McKenzie
Date: 17:06:31 07/17/00
Go up one level in this thread
On July 17, 2000 at 18:25:27, Dann Corbit wrote: >On July 17, 2000 at 18:15:14, Jesus de la Villa wrote: > >>On July 17, 2000 at 13:40:28, Tom Kerrigan wrote: >> >>>I've had a number of requests to implement 3-fold repetition detection in TSCP. >>>It's also clear that TSCP would do better in tournaments (although that isn't >>>the goal...) if it could detect these draws. >>> >>>So the question is, is there an easy way to do the detection? >>> >>>In my "strong" program, I just compare hash keys. But TSCP doesn't keep hash >>>keys and I have no intention for it to do so. So is there another way to do it? >>> >>>Thanks in advance. >>> >>>-Tom >> >>It would be more funny if you put a random factor to return the played move, >>this is with the intention to avoid deterministic play. > >You pretty much have to have an opening book to introduce randomness. If you >just fuzz the eval with random noise, nothing good can come from it. That's >because Alpha-Beta is only going to give you one good move -- you don't know any >of the alternatives. > >If you have an opening book, you can *know* that several alternatives are pretty >good and that (perhaps) 2-3 of them are all very playable. Hence, you can >weight them and introduce randomness. But without an opening book, it will only >generate more losses. > >Or am I wrong. Does anyone really use randomness once you are out of book? In LambChop I used to add a small random number to the eval when searching the first few moves out of book. This was when my book was VERY small (just a few positions) and possibly before I had a proper transposition table, back then it seemed to work OK. I've since taken it out, mainly because I don't need it and because I would be worried about it screwing up my hash table (I don't clear it between moves) for later searches. In summary, adding a small random number to the eval in TSCP might not be such a bad idea. I believe there is some evidence that this can even slightly increases playing strength (seems unlikely I know, I'm not convinced)! Anyway, for TSCP being non-deterministic is probably more important than any change in playing strength. Another safer way to introduce non-determinism is to slightly randomise the thinking time for the first few moves out of book. I do this in LambChop, it works well. cheers, Peter
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.