Author: Robert Hyatt
Date: 13:41:14 12/27/98
Go up one level in this thread
On December 27, 1998 at 15:39:31, Richard A. Fowell (fowell@netcom.com) wrote: >On December 27, 1998 at 10:34:40, Robert Hyatt wrote: > >>On December 26, 1998 at 18:51:46, Richard A. Fowell (fowell@netcom.com) wrote: >> >>>Do hash tables confuse your matefinding routines? >>> >>>I tried this mate in 7: >>>8/krK5/p7/P7/6P1/8/7Q/8 w - - >>> >>>on two different chess programs, by two different authors, >>>and both reported erroneously short mates when their hash tables were on. >>> >>>With hash tables off, the correct solution was given by both programs. > ><snip> > >> >>This has to be a bug. Hash tables can't find a "shorter mate" than actually >>exists on the board, unless there is a bug. Hash tables can let a shallow >>search see "much deeper" than you would expect (position fine#70 is the classic >>example that programs solve at a much shallower depth than is actually >>possible). > >A bug, certainly, but I'm seeing evidence of that bug in three programs, >and one of them is Crafty. > this is always possible. I would suspect a bug in mine/mike's version any time, since we are using parallel search, which is known to be difficult to do and problematic in debugging. There could *easily* be bugs in the current crafty when doing parallel searches... I'll try to study it some more, however, because if it says mate in 8, when there is no mate in 8, then something is definitely wrong. IE I don't do any forward pruning so there's no way I would overlook a move that pushes the mate off for one move, without a bug... >On r.g.c.c., a mate in 9 was posted: >kr6/2K5/p7/P7/6PQ/8/8/8 w - - > >One Crafty report found the mate in 9, >but two evaluated it as mate in 8. Digging through DejaNews, I've found >and appended a few pertinent posts: > >Post #1 - by Robert Hyatt > >You responded to Ed's post that some Crafty users were reporting the mate >in 9, but that Ed's copy of 16.2 reported mate in 8. Ed further pointed >out that the PV given by Crafty could be improved to extend the defence >to mate in 9. It seems you wanted more evidence that it was truly a >mate in 9 before investigating ... you said: > >"But before I say anything the above should be looked at carefully > to be sure there isn't some other unnoticed resource that keeps the > mate correct..." > >Post #2 - Heiner Marxen finds the two mates in 9 using his special-purpose > mate-finding program. > >Post #3 - Mike Byrne reports Crafty finding a mate in 8, > with the other keymove (Qh1+) after > some tweaking of Crafty parameters. > >Post #4 - Heiner Marxen points out an improvement to Black's > defence in the Qh1+ variation, and asks for confirmation > of the mate in 8 (since, if it existed, there would be > an undetected bug in Heiner's program) > >Post #5 - Mike Byrne (four days later) says he added a Crafty mate-finding > routine, and that "this is a mate in 9". I take this to mean > that he confirmed that the mate in 8 was erroneous. He said > he emailed you the matefinding code. > >I ran this problem on Sigma Chess, and it said there was no mate in 8. > >In summary, I think there's confirmation that Crafty's eval of Mat08 >in both Ed Collin's and Mike Byrne's copy of Crafty was erroneous >(per Heiner Marxen, Mike Byrne, and myself), and that Crafty shares this >bug of erroneously short mate evals. Based on my work with Sigma Chess, it >seems that hash tables handling is a likely culprit. > >Since this seems to be a widespread issue, and it seems to affect Crafty, >I was hoping you could help track down what causes this bug, and warn people >about it (heck, as widespread as this seems to be, it might be worth an >(ICCA paper). > >Regards, > >Richard A. Fowell (fowell@netcom.com) > >=============================================== >Edward D. Collins <ecollins@inficad.com> wrote: >:>> kr6/2K5/p7/P7/6PQ/8/8/8 w - - 0 0 > >: Bernhard Bauer wrote... > >:> Crafty16.2 on a 2xPPro 233MHz finds the >:> mate in 9 on ply 11. > >:> The move 1. g5 is found in 6 sec and the >:> whole search is finished in 1:42. > > >:> Here some output from crafty: > >:> 10-> 9.08 9.92 1. g5 Rb7+ 2. Kc6 Rb8 3. g6 Ka7 4. >:> Qh7+ Ka8 5. Qc7 Rb1 6. Qd8+ Rb8 7. >:> Qxb8+ Kxb8 8. g7 Ka7 9. Kd6 >:> 11 10.84 ++ 1. g5!! >:> 11 1:31 Mat09 1. g5 Rb7+ 2. Kc6 Rb8 3. g6 Ka7 4. >:> Qh7+ Ka8 5. g7 Rc8+ 6. Kb6 Rb8+ 7. >:> Kxa6 Rb7 8. g8=Q+ Rb8 9. Qb7# >:> 11-> 1:42 Mat09 1. g5 Rb7+ 2. Kc6 Rb8 3. g6 Ka7 4. >:> Qh7+ Ka8 5. g7 Rc8+ 6. Kb6 Rb8+ 7. >:> Kxa6 Rb7 8. g8=Q+ Rb8 9. Qb7# >:> time=1:42 cpu=200% mat=5 n=30057337 fh=97% nps=292585 >:> ext-> checks=6269540 recaps=290 pawns=277457 1rep=691838 >:> predicted=0 nodes=30057337 evals=126326 >:> endgame tablebase-> probes done=0 successful=0 maxd=32 >:> hashing-> trans/ref=55% pawn=99% used=w89% b99% >:> SMP-> split=2340 stop=534 data=9/64 cpu=3:26 elap=1:42 >:> mate in 9 moves. > > >: Interesting. I hadn't downloaded 16.2 yet so after seeing your post I >: did so. I setup the problem and then let it run while I took a shower >: and had breakfast. > >: I was surprised to discover later that I received different results. In >: my log file, Crafty reported a mate in 8! > >Before you can compare results, all facets of the experiment have to be >the same. You probably didn't use the same hash size, and you definitely >didn't use the parallel search, which brings its own set of quirks into >the game.. > >If anyone is interested, we can certainly discuss "parallel anomalies" >as they can be quite interesting... > > >: 10 30.49 ++ 1. Qh1+!! >: 10-> 32.55 10.27 1. Qh1+ >: 11 32.77 ++ 1. Qh1+!! >: 11 52.23 Mat10 1. Qh1+ Ka7 2. Qh2 Rb7+ 3. Kc8 Rb5 >: 4. Qc7+ Ka8 5. Qc6+ Ka7 6. Kc7 Rb8 >: 7. Qb6+ Rxb6 8. axb6+ Ka8 9. b7+ Ka7 >: 10. b8=Q# >: 11 42:47 Mat08 1. g5 Rb7+ 2. Kd6 Ka7 3. g6 Ka8 4. >: g7 Rxg7 5. Qh8+ Ka7 6. Qxg7+ Ka8 7. >: Kc6 Kb8 8. Qb7# >: 11-> 43:35 Mat08 1. g5 Rb7+ 2. Kd6 Ka7 3. g6 Ka8 4. >: g7 Rxg7 5. Qh8+ Ka7 6. Qxg7+ Ka8 7. >: Kc6 Kb8 8. Qb7# >: time=43:35 cpu=99% mat=5 n=120830485 fh=98% nps=46197 >: ext-> checks=26133140 recaps=653 pawns=1002111 1rep=2880703 >: predicted=0 nodes=120830485 evals=190387 >: endgame tablebase-> probes done=1751199 successful=104850 maxd=32 >: hashing-> trans/ref=47% pawn=13% used=w99% b99% > > >: The thing is, after 5.Qh8+, there is no mate in 8 as Crafty can delay >: the mate by one move by interposing with the Rook (5... Rg8 6.Qxg7 Kb7 >: 7.Qf7+ Kb8 8.Kc6 Ka8 9.Qb7#) > >: Not sure yet why this happened... > >First Crafty isn't supposed to find the "shortest" mate, although it should >never find one shorter than really exists... But before I say anything the >above should be looked at carefully to be sure there isn't some other >unnoticed resource that keeps the mate correct... > > > >: Like I said, I enjoy watching computers work on this one. > > >: - Ed > > >-- >Robert Hyatt Computer and Information Sciences >hyatt@cis.uab.edu University of Alabama at Birmingham >(205) 934-2213 115A Campbell Hall, UAB Station >(205) 934-5473 FAX Birmingham, AL 35294-1170 >======== end of post 1, start of post 2 =========== > >Heiner Marxen >Email: heiner@DrB.Insel.DE >Date: 1998/12/22 >Forums: rec.games.chess.computer > >In article <75nppn$qj6$1@news.inficad.com>, >Edward D. Collins <ecollins@inficad.com> wrote: >>Here's a position I still have fun watching computers attempt to figure >>out. White to play and mate in 9. (I think.) >> >>kr6/2K5/p7/P7/6PQ/8/8/8 w - - 0 0 > >[history snipped] >>However, I think that the optimal solution is a mate in 9 in which there >>are slightly different variations, all beginning with 1.g5! I'll give >>two: >> >>1.g5 Rb7+ 2.Kc6 >> >>a) 2...Rb8 3.g6 Ka7 4.Qe7+ Ka8 5.g7 Rc8+ 6.Kb6 Rb8+ 7.Kax6 Rb6+ 8.Kxb6 >>Kb8 9.g8=Q# > 4.Qd4+ will do also > 3.Qh7 will do also > >Confirmed. > >>b) 2...Kb8 3.Qf4+ Ka8 4.g6 Rb8 5.Qxb8+ Kxb8 6.g7 Ka7 7.Kc7 Ka8 8.g8=Q+ >>Ka7 9.Qb8# > >Confirmed. > >>Try putting this problem on your computer and let me know which solution >>it comes up with and how quickly. > >My chess problem solver (it does *not play* chess) completed this one after >211 seconds (3:31) on a not so fast machine. (No, it's not available.) > >And it came up with a second key move for "mate in 9", which is >our old friend 1.Qh1+. It can be improved, again: > >>a) 1.Qh1+ Ka7 2.Qh2 Rb7+ 3.Kc8 Rb5 [3...Rb6 4.Qc7+ Ka8 5.axb6 a5 6.Qa7#] >>4.Qc7+ Ka8 5.g5 Rc5 6.g6 Rxc7+ 7.Kxc7 Ka7 8.g7 Ka8 9.g8=Q+ Ka7 10.Qb8# > > 4.Qd6 Rb1 5.Qd4+ Ka8 6.Qe4+ Ka7 7.Qxb1 Ka8 8.Qb8# > >Better is 3.... Rb3 (all other moves yield shorter mates) with e.g. > > 4.Qc7+ Ka8 5.Qc6+ Ka7 6.Qc5+ Ka8 7.Qd5+ Ka7 8.Qxb3 Ka8 9.Qb8# > >That was fun! >-- >Heiner Marxen heiner@drb.insel.de http://www.drb.insel.de/~heiner/ > >================= end of post 2, start of post 3 ======= > >Author: Mike Byrne >Email: chessman.nospam@voicetnet.com.nospam >Date: 1998/12/22 >Forums: rec.games.chess.computer > >Crafty is programmed to find mate ...it doesn't really care if one is longer >than the other...but on my dual 400 (without tb's and it is important to set >pondering to *off* before eneterneing the position) ..it does find the >mate-in8 ..every move is forced or the mate is even quicker...with tb's >(kppkp) it finds the *master style*win of >the Qd8 (Mate in 10) > > >White(1): go >end-game phase > clearing hash tables > time surplus 0.00 time limit 1500:00 (1500:00) > depth time score variation (1) > 9 0.28 10.28 1. Qh1+ Ka7 2. Qc6 Rc8+ 3. Kd7 Rxc6 > 4. Kxc6 Kb8 5. g5 Kc8 6. g6 Kd8 > 9-> 1.10 10.28 1. Qh1+ Ka7 2. Qc6 Rc8+ 3. Kd7 Rxc6 > 4. Kxc6 Kb8 5. g5 Kc8 6. g6 Kd8 > 10 1.41 10.32 1. Qh1+ Ka7 2. Qc6 Rc8+ 3. Kd7 Rxc6 > 4. Kxc6 Kb8 5. g5 Kc8 6. g6 Kd8 7. > g7 Ke7 > 10 4.02 ++ 1. g5!! > 10-> 4.97 10.71 1. g5 > 11 5.71 ++ 1. g5!! > 11 29.22 Mat10 1. g5 Rb7+ 2. Kd6 Rb8 3. g6 Ka7 4. > Qh7+ Ka8 5. g7 Rc8 6. g8=R Rxg8 7. > Qxg8+ Ka7 8. Qg7+ Ka8 9. Kc6 Kb8 10. > Qb7# > 11 33.16 Mat08 1. Qh1+ Ka7 2. Qh2 Rb7+ 3. Kc8 Rb5 > 4. Qd6 Rb1 5. Qe7+ Ka8 6. Qe4+ Ka7 > 7. Qxb1 Ka8 8. Qb8# > 11-> 34.21 Mat08 1. Qh1+ Ka7 2. Qh2 Rb7+ 3. Kc8 Rb5 > 4. Qd6 Rb1 5. Qe7+ Ka8 6. Qe4+ Ka7 > 7. Qxb1 Ka8 8. Qb8# > time=34.22 cpu=198% mat=5 n=16118704 fh=96% nps=471031 > ext-> checks=3099819 recaps=136 pawns=177186 1rep=507936 > predicted=0 nodes=16118704 evals=80119 > endgame tablebase-> probes done=0 successful=0 maxd=32 > SMP-> split=2080 stop=432 data=9/64 cpu=1:08 elap=34.22 > > > >Bernhard Bauer wrote in message <367FA4C2.590128AC@iag.uni-stuttgart.de>... >> >> >> >> >> >>Crafty16.2 on a 2xPPro 233MHz finds the mate in 9 on ply 11. >>The move 1. g5 is found in 6 sec and the whole search is finished in 1:42. >>Here some output from crafty: >> >> 10-> 9.08 9.92 1. g5 Rb7+ 2. Kc6 Rb8 3. g6 Ka7 4. >> Qh7+ Ka8 5. Qc7 Rb1 6. Qd8+ Rb8 7. >> Qxb8+ Kxb8 8. g7 Ka7 9. Kd6 >> 11 10.84 ++ 1. g5!! >> 11 1:31 Mat09 1. g5 Rb7+ 2. Kc6 Rb8 3. g6 Ka7 4. >> Qh7+ Ka8 5. g7 Rc8+ 6. Kb6 Rb8+ 7. >> Kxa6 Rb7 8. g8=Q+ Rb8 9. Qb7# >> 11-> 1:42 Mat09 1. g5 Rb7+ 2. Kc6 Rb8 3. g6 Ka7 4. >> Qh7+ Ka8 5. g7 Rc8+ 6. Kb6 Rb8+ 7. >> Kxa6 Rb7 8. g8=Q+ Rb8 9. Qb7# >> time=1:42 cpu=200% mat=5 n=30057337 fh=97% nps=292585 >> ext-> checks=6269540 recaps=290 pawns=277457 1rep=691838 >> predicted=0 nodes=30057337 evals=126326 >> endgame tablebase-> probes done=0 successful=0 maxd=32 >> hashing-> trans/ref=55% pawn=99% used=w89% b99% >> SMP-> split=2340 stop=534 data=9/64 cpu=3:26 elap=1:42 >> >>mate in 9 moves. >> ><snip> >> >>Kind regards >>Bernhard >>"Edward D. Collins" wrote: >> >>> Here's a position I still have fun watching computers attempt to figure >>> out. White to play and mate in 9. (I think.) >>> >>> kr6/2K5/p7/P7/6PQ/8/8/8 w - - 0 0 >>> >>> If memory serves, this was originally published/presented as a mate in >>> 15 with the Queen aesthetically zigzagging up the board... >>> >>> 1.Qh1+ Ka7 2.Qg1+ Ka8 3.Qg2+ Ka7 4.Qf2+ Ka8 5.Qf3+ Ka7 6.Qe3+ Ka7 7.Qe4+ >>> Ka8 >>> 8.Qd4+ Ka8 9.Qd5+ Ka7 10.Qc5+ Ka8 11.Qc6+ Ka7 12.Qb6+ Rxb6 13.axb6+ Ka8 >>> 14.b7+ >>> Ka7 15.b8=Q# >>> >>> ... which many programs will still find first, since every move is a >>> check. >>> >>> However, someone, somewhere, actually discovered a two move shorter mate >>> beginning with the non checking move 9.Qd6 >>> >>> 1.Qh1+ Ka7 2.Qg1+ Ka8 3.Qg2+ Ka7 4.Qf2+ Ka8 5.Qf3+ Ka7 6.Qe3+ Ka7 7.Qe4+ >>> Ka8 >>> 8.Qd4+ Ka8 9.Qd6 Rb7+ 10.Kc8 Ka7 11.Qc5+ Ka8 12.Qc6 Ka7 13.Qxb7# >>> >>> Not to be outdone, someone THEN shaved off three more moves from the >>> solution and discovered that at some a mate in 10 was possible... >>> actually a couple of them, each beginning with the subtle 2.Qh2. >>> >>> a) 1.Qh1+ Ka7 2.Qh2 Rb7+ 3.Kc8 Rb5 [3...Rb6 4.Qc7+ Ka8 5.axb6 a5 6.Qa7#] >>> 4.Qc7+ Ka8 5.g5 Rc5 6.g6 Rxc7+ 7.Kxc7 Ka7 8.g7 Ka8 9.g8=Q+ Ka7 10.Qb8# >>> >>> b) 1.Qh1+ Ka7 2.Qh2 Rb7+ 3.Kc8 Rb5 4.Qc7+ Ka8 5.Qc6+ Ka7 6.Kc7 Rb1 >>> 7.Qb6+ >>> Rxb6 8.axb6+ Ka8 9.b7+ Ka7 10.b8=Q# >>> >>> However, I think that the optimal solution is a mate in 9 in which there >>> are slightly different variations, all beginning with 1.g5! I'll give >>> two: >>> >>> 1.g5 Rb7+ 2.Kc6 >>> >>> a) 2...Rb8 3.g6 Ka7 4.Qe7+ Ka8 5.g7 Rc8+ 6.Kb6 Rb8+ 7.Kax6 Rb6+ 8.Kxb6 >>> Kb8 9.g8=Q# >>> >>> b) 2...Kb8 3.Qf4+ Ka8 4.g6 Rb8 5.Qxb8+ Kxb8 6.g7 Ka7 7.Kc7 Ka8 8.g8=Q+ >>> Ka7 9.Qb8# >>> >>> Try putting this problem on your computer and let me know which solution >>> it comes up with and how quickly. >>> >>> - Ed >> >================ end of post 3, start of post 4 =========== > >Author: > > Heiner Marxen >Email: > heiner@DrB.Insel.DE >Date: > 1998/12/22 >Forums: > rec.games.chess.computer >more headers > > > > author profile > > view thread > > > >In article <1JQf2.500$7P2.6731730@news2.voicenet.com>, >Mike Byrne <chessman.nospam@voicetnet.com.nospam> wrote: >[snip] >> 11-> 34.21 Mat08 1. Qh1+ Ka7 2. Qh2 Rb7+ 3. Kc8 Rb5 >> 4. Qd6 Rb1 5. Qe7+ Ka8 6. Qe4+ Ka7 >> 7. Qxb1 Ka8 8. Qb8# > >According to my program, 3... Rb3 cannot be completed as Mat08. >If crafty is right, my program is wrong <sniff>. So, could you >please ask your crafty to show a solution to that move? > >See also my direct followup. >-- >Heiner Marxen heiner@drb.insel.de http://www.drb.insel.de/~heiner/ >============ end of post 4, start of post 5 ======= > >Author: Mike Byrne >Email: @voicetnet.com.nospam >Date: 1998/12/26 >Forums: rec.games.chess.computer > >I just added a new mate finder command for crafty <which I have passed on >Bob Hyatt> > >this is a mate in 9: > > 18-> 6:41 Mat09 1. g5 Rb7+ 2. Kc6 Rb8 3. g6 Ka7 4. > Qe7+ Ka8 5. g7 Rc8+ 6. Kb6 Rb8+ 7. > Kxa6 Rb6+ 8. axb6 Kb8 9. g8=Q# >time=6:41 cpu=199% mat=5 n=205570273 fh=98% nps=512324 >ext-> checks=30969436 recaps=1748 pawns=3984818 1rep=8931601 >predicted=0 nodes=205570273 evals=247991 >endgame tablebase-> probes done=0 successful=0 maxd=32 >SMP-> split=7122 stop=451 data=10/64 cpu=13:20 elap=6:41 >Mike Byrne wrote in message ... >>> 11-> 1:42 Mat09 1. g5 Rb7+ 2. Kc6 Rb8 3. g6 Ka7 4. >>> Qh7+ Ka8 5. g7 Rc8+ 6. Kb6 Rb8+ 7. >>> Kxa6 Rb7 8. g8=Q+ Rb8 9. Qb7# >>> time=1:42 cpu=200% mat=5 n=30057337 fh=97% nps=292585 >>> ext-> checks=6269540 recaps=290 pawns=277457 1rep=691838 >>> predicted=0 nodes=30057337 evals=126326 >>> endgame tablebase-> probes done=0 successful=0 maxd=32 >>> hashing-> trans/ref=55% pawn=99% used=w89% b99% >>> SMP-> split=2340 stop=534 data=9/64 cpu=3:26 elap=1:42 >>> >>>mate in 9 moves. >>> >>> >>>Kind regards >>>Bernhard >>>"Edward D. Collins" wrote: >>> >>>> Here's a position I still have fun watching computers attempt to figure >>>> out. White to play and mate in 9. (I think.) >>>> >>>> kr6/2K5/p7/P7/6PQ/8/8/8 w - - 0 0 >>>> > >>>> >>>> However, I think that the optimal solution is a mate in 9 in which there >>>> are slightly different variations, all beginning with 1.g5! I'll give >>>> two: >>>> >>>> 1.g5 Rb7+ 2.Kc6 >>>> >>>> a) 2...Rb8 3.g6 Ka7 4.Qe7+ Ka8 5.g7 Rc8+ 6.Kb6 Rb8+ 7.Kax6 Rb6+ 8.Kxb6 >>>> Kb8 9.g8=Q# >>>> >>>> b) 2...Kb8 3.Qf4+ Ka8 4.g6 Rb8 5.Qxb8+ Kxb8 6.g7 Ka7 7.Kc7 Ka8 8.g8=Q+ >>>> Ka7 9.Qb8# >>>> >>>> Try putting this problem on your computer and let me know which solution >>>> it comes up with and how quickly. >>>> >>>> - Ed >>> >> >>
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.