Author: Miguel A. Ballicora
Date: 07:54:15 12/18/00
Go up one level in this thread
On December 18, 2000 at 09:39:19, Tim Foden wrote: >On December 18, 2000 at 07:46:55, Ulrich Tuerke wrote: > >>Forget my remarks. If this was your problem, you would probably not reach so >>high iterations. (Comet ususally solves this on iteration #18.) >>I guess that rather your algo has a bug. > >I think you may well be right. I could try stripping *everything* out of the >search and the evaluation function, except the hash table and material >evaluation. > >Should the FINE 70 position still be solved with such a search? I think not! I made that mistake and I spent some time trying to find a bug that apparently did not exist (or I did not find :-) I sent a long post several days ago (more than a week ago) about it. I found that the efficieny of the hashing depended a lot on the evaluation. I don't know if this is crazy but it is what I found. If there is any comment about that, I will appreciate it. Do you have that post? I can post it again if you like. I would try this evaluation function: 1) If a black pawn is not in a5 or f5 ---> bonus = 2.0 (so you'll see exactly when the program finds the capture) 2) add a little bonus (~0.01) to the evaluation according to the rank where the king is (and also black king but this was not relevant). I believe that this is important because it drives the king towards the opponent field. I found that including "2" in the evaluation help tremendously in the hashing. I had no apparent bug, but using a better replacement scheme was less dependent on the the evaluation. On the other hand, use 1) to detect when the program *sees* the capture because I found that there are variations that the program sees it but I did not realize it. Those variations where when White king took the pawn in a5 and BK took the one in f4! with only material evaluation it took several plies and lots of nodes so I could see an increase in the score!!! If you use 1 and 2 we can compare the outputs. Do not forget to disable null move. Miguel > >> >>Sorry, Uli > >No problem... any ideas are appreciated. :) > >Cheers, Tim.
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.