Author: Vincent Diepeveen
Date: 14:26:23 04/28/04
Go up one level in this thread
On April 28, 2004 at 15:03:41, Gerd Isenberg wrote: >On April 28, 2004 at 12:12:17, Vincent Diepeveen wrote: > >>On April 28, 2004 at 11:32:00, Gerd Isenberg wrote: >> >>>On April 28, 2004 at 09:44:01, Vincent Diepeveen wrote: >>> >>>>>I tried double nullmove suggested by Vincent too, without any precondition like >>>>>eval - margin >= beta. Vincent is probably right - it seems to work great, and >>>>>solves a lot of test positions a few percent faster. But with some positions, >>>> >>>>And how many plies does double nullmove on average search deeper in a normal >>>>chess position? >>> >>>It looks like iterations terminate about 5-10% earlier on average. >>>I fear my BF isn't that good to earn an additional ply in standard matches. >>> >>>My impression with your really great double nullmove idea is >>>that it doesn't work so good in critical and probably game decisive positions, >>>even in early middlegame root positions. But of course that may be due to my >>>implemetation and sideeffects with other things in my search. >> >>this is a big nonsense and you know it. > >Sorry Vincent, no - simply my vague impression. >Similar to some more additional knowledge, which slows you down a bit, but helps >to find some moves one ply or even more plies earlier. > >Of course i have to investigate it a bit more and play some matches. > >> >>theoretical chess studies which never come onto the board you cannot call >>'critical game deciding positions'. >> >>You show up with some insane positions where uncommon passer knowledge, checks >>or pawn evaluation are decisive in terms of whether you find a winning move or >>not and you use that to test 'zugzwang'. > >I thought BT-test positions are from some gm matches? >Anyway, there are some other positions i tried so far from my favourite testset, >even positions from games IsiChess played, like this am position. > >6k1/4bppp/2p3n1/5Q2/1qPB4/1P1R1BP1/r6P/7K w - - am Qc8 Another useless position. nullmove has nothing to do with it. Basically you want to avoid a draw score (in fact it's losing) for the line shown by diep at 12 ply: 00:03 196987 701276 0 (2) 8 (18,149) 0.157 Bd4-g1 Ra2-a1 Qf5-e4 Qb4-a5 Rd3-d1 Ra 1xd1 Bf3xd1 Ng6-e5 00:07 207529 1510818 0 (2) 9 (47,269) 0.596 Bd4-g1 Ra2-a1 Rd3-d1 Ra1xd1 Bf3xd1 N g6-f8 Qf5-e4 Be7-c5 Bd1-c2 Bc5xg1 Kh1xg1 ++ f5-c8 procnr=1 terug=597 org=[596;597] newwindow=[596;520000] 00:08 208348 1848052 0 (2) 9 (63,315) 0.831 Qf5-c8 Ng6-f8 Qc8xc6 Qb4-e1 Bd4-g1 R a2-a1 Qc6-b6 Qe1-f1 Bf3-e4 f7-f5 Be4-d5 Kg8-h8 00:11 207726 2428323 0 (2) 10 (96,514) 0.831 Qf5-c8 Ng6-f8 Qc8xc6 Qb4-e1 Bd4-g1 Ra2-a1 Qc6-b6 Qe1-f1 Bf3-e4 f7-f5 Be4-d5 Kg8-h8 00:24 210851 5123688 0 (2) 11 (147,812) 0.617 Qf5-c8 Be7-f8 Qc8-e8 Qb4-a3 Qe8xc6 Qa3-c1 Bf3-d1 Qc1-b1 Qc6-f3 Bf8-b4 Qf3-f1 f7-f6 01:21 210359 17064390 0 (2) 12 (359,1731) 0.000 Qf5-c8 Ng6-f8 Qc8xc6 Qb4-e1 Bd4- g1 Ra2-a1 Qc6-b6 Qe1-f1 Bf3-e4 Qf1-e2 Be4-f5 Qe2-f1 Bf5-e4 ++ d4-g1 procnr=1 terug=1 org=[0;1] newwindow=[0;520000] 02:59 212145 38090694 0 (2) 12 (384,1821) 0.427 Bd4-g1 Ra2-a1 Rd3-d1 Ra1xd1 Bf3x d1 Ng6-f8 Bd1-c2 g7-g6 Qf5-f3 Be7-c5 Bg1xc5 Qb4xc5 Bc2-e4 f7-f5 All you need is a mate in 1 extension to solve this position quicker. Diep does not have it, so it needs a stupid 12 ply to find it. Let me put diep to Adaptive nullmove R=3,R=2 (if depthleft <= 4 ply left i use R=2) to proof my point: 00:13 88936 1188185 0 (2) 8 (21,152) 0.157 Bd4-g1 Ra2-a1 Qf5-e4 Qb4-a5 Rd3-d1 R a1xd1 Bf3xd1 Ng6-e5 00:23 85233 1961218 0 (2) 9 (55,268) 0.596 Bd4-g1 Ra2-a1 Rd3-d1 Ra1xd1 Bf3xd1 N g6-f8 Qf5-e4 Be7-c5 Bd1-c2 Bc5xg1 Kh1xg1 ++ f5-c8 procnr=1 terug=597 org=[596;597] newwindow=[596;520000] 00:26 89766 2389595 0 (2) 9 (64,299) 0.831 Qf5-c8 Ng6-f8 Qc8xc6 Qb4-e1 Bd4-g1 R a2-a1 Qc6-b6 Qe1-f1 Bf3-e4 f7-f5 Be4-d5 Kg8-h8 00:42 81927 3454893 0 (2) 10 (92,503) 0.831 Qf5-c8 Ng6-f8 Qc8xc6 Qb4-e1 Bd4-g1 Ra2-a1 Qc6-b6 Qe1-f1 Bf3-e4 f7-f5 Be4-d5 Kg8-h8 01:07 83603 5615629 0 (2) 11 (150,798) 0.625 Qf5-c8 Ng6-f8 Qc8xc6 Qb4-e1 Bd4-g1 Qe1-b1 Qc6-e4 Be7-b4 Rd3-e3 Qb1-c1 Bf3-e2 Bb4-c5 02:51 135608 23258259 0 (2) 12 (227,1245) 0.001 Qf5-c8 Ng6-f8 Qc8xc6 Qb4-e1 Bd4- g1 Ra2-a1 Qc6-b6 Nf8-e6 Rd3-d7 Be7-f8 Bf3-d5 Ne6-c5 Bd5xf7 Kg8-h8 ++ d4-g1 procnr=0 terug=2 org=[1;2] newwindow=[1;520000] 05:27 169710 55529294 0 (2) 12 (246,1343) 0.449 Bd4-g1 Qb4-b8 Rd3-d7 Ra2-a1 Rd7- d1 Ra1-a3 Bf3xc6 Ng6-e5 Bc6-a4 Qb8-b7 Rd1-d5 Ne5xc4 It finds of course this tactical trick at the same depth with more nodes. don't watch the search times. shared memory f'ed up a bit again despite that i virtuallock() it. Windows XP is one big unsecure bug. I bet forward pruning works like the sun in this position. You just want to hunt a few attacked pieces here. >> >>Both your positions have nothing to do with zugzwang in absolute respect. Just >>get a fullwidth search with singular extensions and in your testset positions it >>will outgun anything you do with nullmove, i'm sure of it. > >Ok, i will try and see. > >> >>As long as we are not speaking at the same level with each other, i can't help >>it that fullwidth+singular extensions solve anything you show up with sooner. >> > >I fear with search topics and parallel search i'll never reach your level and >brilliancy ;-) > >Cheers, >Gerd
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.