Author: Dann Corbit
Date: 17:14:26 01/14/02
Go up one level in this thread
On January 12, 2002 at 12:42:19, Tony Werten wrote: >On January 11, 2002 at 14:27:47, Dann Corbit wrote: > >>On January 11, 2002 at 13:00:46, Miguel A. Ballicora wrote: >> >>>On January 11, 2002 at 12:47:29, David Rasmussen wrote: >>> >>>>On January 11, 2002 at 12:40:03, Matthias Gemuh wrote: >>>> >>>>>On January 11, 2002 at 11:44:39, David Rasmussen wrote: >>>>> >>>>>>[D]2Q2n2/2R4p/1p1qpp1k/8/3P3P/3B2P1/5PK1/r7 w - - 0 1 >>>>>> >>>>>>In the above position, my search tree explodes because of extensions. It takes > >>>>>>3 minutes and 42 million nodes to finish an 8 ply search. How does your program >>>>>>do? >>>>>> >>>>>>Any good ideas on how to limit extensions in such a position. >>>>>> >>>>>>P.S. Solving the position is no problem. It is a simple mate in 5 plies, and is >>>>>>found very early on at depth 1 or 2 or so. Still, there must be something >>>>>>unsound about my extensions, or at least room for improvement, when this >>>>>>position makes the tree explode. >>>>>> >>>>>>/David >>>>> >>>>> >>>>> >>>>> >>>>>My HyLogicChess has the following stats: >>>>>All extensions active. >>>>>Quiescence moves: 29% (no explosion, my average is normally have 40%) >>>>>Total moves searched: 11000 >>>>>Speed; 84 kN/s >>>>>Move Ordering: 78% (normally 90%..92%) >>>>>[1] (0.18) 1.f3 (00h:00m:00s) >>>>>[2] (0.12) 1.Qb7 e5 (00h:00m:00s) >>>>>[3] (14.50) 1.Qxf8 Kh5 2.Rxh7 Kg4 3.Qxd6 (00h:00m:00s) >>>>>[4] (+Mate in 3) 1.Qxf8 Kh5 2.Rxh7 Kg4 3.f3 (00h:00m:00s) >>>> >>>>My program performs similarly if only searching to ply 4, but try letting it >>>>search to ply 9, even though it has already found the mate. That is what I am >>>>talking about. >>>> >>>>/David >>> >>>Gaviota's output: >>> >>>setboard 2Q2n2/2R4p/1p1qpp1k/8/3P3P/3B2P1/5PK1/r7 w - - 0 1 >>>d >>>+-----------------+ >>>| . . Q . . n . . | >>>| . . R . . . . x | >>>| . x . q x x . k | >>>| . . . . . . . . | >>>| . . . o . . . o | >>>| . . . B . . o . | >>>| . . . . . o K . | >>>| r . . . . . . . | >>>+-----------------+ >>> >>>sd 9 >>>go >>> 104 1: 0.0 +30.78 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8xd6 >>> 392 2: 0.0 +30.78 Qc8xf8 Kh6-h5 Rc7xh7 <EMPTY> >>> <-transp >>> 455 3 0.0 :-) Qc8xf8 >>> 530 3 0.0 :-) Qc8xf8 >>> 565 3 0.0 :-) Qc8xf8 >>> 1637 3 0.0 :-) Qc8xf8 >>> 3104 3: 0.0 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 4023 4 0.1 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 4475 4: 0.1 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 5444 5 0.1 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 8853 5: 0.1 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 11891 6 0.1 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 18148 6: 0.2 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 24692 7 0.2 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 48189 7: 0.4 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 62166 8 0.6 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 152926 8: 1.5 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 172449 9 1.6 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> 262305 9: 2.3 +Mat_3 Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 >>> Qf8-g8 >>> >>>Ply: 9 >>>Qc8xf8 Kh6-h5 Rc7xh7 Kh5-g4 Qf8-g8 >>>Score: 125.37 (32094) Evals: 90689 Time: 2.3s nps: 115400 Q/all: 0.36 >>> nodes cutoffs missed tree_exp >>>path 168830 26576 823 1.03 >>>quies 93475 31592 1083 1.03 >>>all 262305 58168 1906 1.03 >>>hashtable= attempts: 284553 hits: 35.3% perfect: 23.2% >>>move c8f8 >> >>For a mate in 3, it does not make any sense to search past ply 5 if you already >>see it. After all, when you have searched ply 5 and have a mate in 3, you >>cannot possibly improve it. > >Yes you can. Nullmove and pruning can take care of hiding a short mate until >deeper searchdepth. > >The safe assumption is that when you have a mate in 5 ply and you're deeper than >4 ply in the search you can cut off since it will not improve the score. That's the same thing I said. A mate in 3 is a mate in 6 ply.
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.