Author: Vincent Diepeveen
Date: 14:30:56 08/20/02
Go up one level in this thread
On August 20, 2002 at 16:05:17, David Hanley wrote: This is a sick statement David and i hope you realize it. You can't risk randomness simply in a program. I remember a draughts game where i drew with Napoleon. It missed a very obvious win 3 vs 1 (this was before EGTBs were there). When i debugged it the same evening at home, back from the tournament, i directly found out without hashtable i found the 3 vs 1 win instantly instead of letting the opponent draw the game. Then i debugged and debugged all night in order to find out it was a collission which caused it. I stored in Napoleon 32 bits and indexed with 20 bits or so. Then i did an additional 8 probes (so reducing that 20 bits by at least 3, but practical 4). So we talk about an effective hashkey length of about 48 bits. It was the reason why i started testing it in DIEP too. I then stored 44 bits (most significant 40 bits and least significant 4 bits) and it didn't fail to find the 3 vs 1 win. Now you are prepared to take the risk of this scenario? Long after you have forgotten all this you will still play awful with your program. Actually i measued for diep that 'only' once each 200MLN positions i had a very *bad* collission causing the tree to get a wrong score, but usually of course not backed up to the root. Just calculate with me: a) opponent plays to a position favourable for it b) your thing doesn't care. At say 7 tournament games with 60 moves each game, what is the chance that you play towards such a position? Right, *very* big. >I read with interest the experiments with crafty with introducing hash false >hits artifically into the search tree. I recall the experiments showed that >adding the false hits, even at a very high rate 1 per 1000, had little of no >effect on the search results. > >It would be good for me to use 32-bit signatures in my program, and i'm >wondering if the above result indicates that 32-bit signatures won't matter so >much in my program, which is a slow searcher. Certinaly if i check hamming >distance, 32 bit hash false matches should be at a far lower rate than the one >per thousand that didn't adversely affect crafty. > >dave
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.