Author: leonid
Date: 07:46:30 03/11/01
Go up one level in this thread
On March 10, 2001 at 20:59:45, Heiner Marxen wrote: >On March 10, 2001 at 17:26:33, leonid wrote: > >>On March 10, 2001 at 16:04:47, Heiner Marxen wrote: >> >>>On March 10, 2001 at 08:11:29, leonid wrote: >>> >>>>Hi! >>>> >>>>If you like to solve one light mate then you can try this: >>>> >>>>[D]nqNBNqn1/2QKQ3/Nq3qN1/n1rkr1n1/R1NbN1R1/1n1q1n2/8/Q2B2Q1 w - - >>>> >>>>Please, indicate your result. >>>> >>>>Thanks, >>>>Leonid. >>> >>>Chest needs just 71 seconds (K7/600) and mere 16MB hash to find 8 (!) >>>solutions in 8 moves. Here are the PVs: >>> >>>N4xb6+ Nxb6+ Nxb6+ Qfxb6 Qgxd4+ Qxd4 Qxd4+ Nbxd4 N8f6+ Nxf6+ Nxf6+ Qfxf6 >>>Qexc5+ Qxc5 Qxc5# >>>N4xf6+ Nxf6+ Nxf6+ Qbxf6 Qgxd4+ Qxd4 Qxd4+ Nfxd4 N8b6+ Nxb6+ Nxb6+ Qbxb6 >>>Qexe5+ Qxe5 Qxe5# >>>N8xb6+ Nxb6+ Nxb6+ Qfxb6 Qgxd4+ Qxd4 Qxd4+ Nbxd4 N8f6+ Nxf6+ Nxf6+ Qfxf6 >>>Qexc5+ Qxc5 Qxc5# >>>N8xf6+ Nxf6+ Nxf6+ Qbxf6 Qgxd4+ Qxd4 Qxd4+ Nfxd4 N8b6+ Nxb6+ Nxb6+ Qbxb6 >>>Qexe5+ Qxe5 Qxe5# >>>Nc3+ Bxc3 N4xb6+ Qbxb6 Bxb3+ Nxb3 Qgxc5+ Nxc5+ Qexc5+ Qbxc5 Nf4+ Qxf4 >>>Qb7+ Qc6+ Qxc6# >>>Ne3+ Bxe3 N4xf6+ Qfxf6 Bxf3+ Nxf3 Qaxe5+ Nxe5+ Qcxe5+ Qbxe5 Qh1+ Qde4 >>>Qxe4+ Qxe4 Qxe4# >>>Qaxd4+ Qxd4 N8xf6+ Nxf6+ Nxf6+ Qbxf6 Qxd4+ Nfxd4 N8b6+ Nxb6+ Nxb6+ Qbxb6 >>>Qexe5+ Qxe5 Qxe5# >>>Qgxd4+ Qxd4 N4xf6+ Nxf6+ Nxf6+ Qbxf6 Qxd4+ Nfxd4 N8b6+ Nxb6+ Nxb6+ Qbxb6 >>>Qexe5+ Qxe5 Qxe5# >>> >>>Here Chest is comparatively fast, since black can attack the white king, >>>once white does a non-checking move. >>> >>>Heiner >> >>Hi, Heiner! >> >>Your time is very nice and close to what I could see. So, very often (not all >>the time) good branching factor is indicative for very solvable position. I was >>not wrong when indicated this position as "light". Mine took by brute force in >>108 sec. on AMD 400. > >Hey! That is nearly exactly the same solution time, if scaled by the MHz! > > >>And what was average NPS for this position is you looked into it by brute force. >>I want to see where is my brute force NPS. I have now this counter only for >>brute force but not for selective. For selective search expect to install later. >>For brute force my NPS goes around 6 or 7 times below of what I expected after >>other past of my chess program. For this position (mate solver) NPS is "very >>high". It is around 100K on AMD 400Mhz. >> >>Cheers, >>Leonid. > >Chest made/unmade 4495004 moves during 71.13 seconds: 63194 N/s >That may appear a bit low, but Chest tries to avoid move execution near >the leaves, and tries to prove their uselessness. Therefore, NPS is low, >but "time to depth" is good. This is a bit like a "slow searcher". Now I see that our NPS is practically the same. This very funny! I found many time that programs that have nothing to do in its inner logic end somewhere by the same general figures. Only before I could see this for chess programs. Ha, Heiner, I remember that I have seen NPS for solving the mate for Rebel 10. Its numbers are impressive! Later still will look for some concret position but it should be expected around 200k on my AMD 400Mhz. >Regarding your unexpectedly low NPS: detecting a mate (the eval function of >a mate searcher/prover) can be a lot of work. It depends on what data is >present, and what you have told over the past, I guess that your program >does not update much more than the bare minimum of data. That may be (part of) >an explanation. Until last two weeks my solved never looked into last experience. Only around more that one week ago I went to change it slighly for putting into it "previous best move" for obtaining better move aligning. It became now slighly better, but change is not spectacular. The best result is around 25% for brute force search in 8 moves deep search. Expect to go to my mate solver later. Talking here about mate solution help in doing this. Leonid. >Nevertheless, on this problem your program was really very good! > >Heiner
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.