Author: José Carlos
Date: 11:40:38 01/08/01
Go up one level in this thread
On January 08, 2001 at 08:21:50, Severi Salminen wrote: >>>Congratulations, doesn't it feel good! It was me who "first" beat TSCP. One >>>question (I'm not familiar with terminology): what is "Hash table for refutation >>>and transposition"? Is this the normal hash table in all programs (not in mine >>>yet). I understand the "transposition", but what is "refutation"? >>> >>>And have you tried your program's evaluation without mobility factor? It should >>>make it faster and maybe without affecting the precision of your eval(). Try it >>>and if possible make 200 or more games (verison with mobility vs. version >>>without) and report here at CCC. It would be very interesting to see results! >> >> I have mobility in Averno and, when I take it out, the program seems to become >>instantly stupid. Actually, it's much weaker (~150 ELO) without mobility in its >>eval. Maybe I'd need a better overall eval to be able to take mobility out, I >>don't know, but I can assure that, right now, I can't live without mobility :) > > >Very interesting. When you evaluate a position at leaf node do you have to >generate all possible moves for both sides? I do not generate all moves. I have a special code that counts them. >This sounds very time consuming. Or >do you have a "secret" trick to do this faster? No secret trick at all. Actually my program is open-source :) When I first wrote mobility routine, I did generate all moves, then substract last minus first and then discard them all. It was really stupid. :) Later, I wrote some code that does something similar, but only counts (so I save the call to addmove and all that stuff). The make-unmake part in my program is much slower than move generation (mainly because of inCheck()), so it's not really so important the loss of time doing mobility, compared to the gain in positional play. >And are you using andy >piece-square evaluation in addition? I have piece-square, but it's not enough. The program tries to put their pieces in some squares and then, ok, it is happy with the position, no need to do anything else :) It's not exactly that way, but believe me, the program looks absolutly stupid without mobility. >If you _don't_ have piece-square evaluation >then it is very understandable that your program becomes stupid when mobility is >removed. > >Severi Mobility was the first thing I implemented (after material), because it sounded very reasonable to me that, the more moves you can do, the more probability you have to find a good one. I recommend you an experiment: implement mobility, and take all the rest of eval out (exept material). I bet you'll be surprised of how logical moves mobility alone makes. José C.
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.