Author: Stuart Cracraft
Date: 19:26:28 07/04/04
Go up one level in this thread
On July 03, 2004 at 17:19:39, Uri Blass wrote: >On July 03, 2004 at 14:45:24, Stuart Cracraft wrote: > >>On July 03, 2004 at 07:12:24, David B Weller wrote: >> >>>Just a thought - >>> >>>When the reduced depth verification search fails high, what depth do you store >>>in hash? I think it should be original depth, right? Not the reduced depth. >>> >>>David >> >>I store the original depth in the hashtable after failed failhigh in >>null move verification. >> >>On the other hand, after extensions like pawn to 6th/7th, recapture, >>and the like, I store the increased depth in the hashtable. >> >>Hope this isn't too much against the grain. Still have a ton of bugs. >> >>Solve 152 out of 300 of the Win-at-Chess in 1 second though with >>everything enabled. If I disable null move verification I get 167 >>out of 300 at 1 second. >> >>Wonder if null move has this tactical-reducing capability in other >>programs? -- So that I know if it is working through indirect evidence >>(besides direct evidence of depth, nodes, etc.) >> >>Anyone care to comment about that last question? > >I think that if you solve 152 or even 167 out of 300 of the win in chess in 1 >second then you have clearly bigger problem than null move pruning. > >Tscp does not support FEN but I guess that even tscp that has no null and no >hash can get more than 167 out of 300 in one second. I'm the first to admit I have a ton of bugs. As Heinz said, every program is a fulcrum of discovery for improvement... > >Did you test that pawn to 6/7 extensions help you? >I do not extend pawn to the 6th. I tried both. No extensions helped in a 300 position test suite. Tried recapture, check extension, 6th/7th pawn push, 7th pawn push only. Also tested null move. None of these helped produce a better score. Only transposition table. > >I think that you did the mistake of implementing a lot of stuff without testing. >You should implement one thing at a time and test if it is productive. Always do this. > >I suggest that you delete null move and hash and extensions and test again and >only implement things that help you to score better in WAC. > Of course, the situation arises that my implementation may not be good -- but I check various sites/programs for comparison (Moreland, Schroder, GNU, TSCP, etc.)... from a tactical test-problem standpoint, these extensions or at least my implementation, did not impress. Null move reducing tactics slightly is well-known. I did not want to get bogged down in a quagmire of algorithms again with this implementation, wanting only something clean and quick that wasn't someone else's code for the real phase of the research. Alas, there seems no way not to get bogged down with the algorithmic quicksand and gordian knot of testing. Stuart
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.