Author: F. Huber
Date: 07:15:06 01/07/05
Go up one level in this thread
On January 07, 2005 at 09:37:55, Heiner Marxen wrote: >On January 06, 2005 at 12:59:29, F. Huber wrote: > >>On January 06, 2005 at 12:20:13, Heiner Marxen wrote: >> >>>On January 05, 2005 at 16:15:22, F. Huber wrote: >>> >>>>On January 05, 2005 at 14:02:58, Richard Pijl wrote: >>>> >>>>>On January 05, 2005 at 13:47:09, Kurt Utzinger wrote: >>>>> >>>>>>Much too easy for any chess program I think. Or can you find >>>>>>a modern engine being unable to solve this within 0 to some seconds? >>>>>>Kurt >>>>> >>>>>I only tried a few, and all were able to solve it in 1 second or less on my >>>>>machine. >>>>>Now a question for the mate-solvers: Is this a mate in 20, or can a faster mate >>>>>be found? >>>>>Richard. >>>> >>>>Hi Richard, >>>> >>>>I´ve tried it now in brute-force mode with ´Gustav 3.0´ up to depth 13, but >>>>with a branching factor of about 2.5 it would take too long, if it´s really >>>>a mate in 20 (so at least it´s above 13). >>>>Chest (also in brute force) is much slower for this position ... >>>> >>>>Of course all this is _without_ any EGTBs, since both programs don´t support >>>>them - although: Heiner Marxens private development version of Chest seems to >>>>already have implemented it, so maybe Heiner could try this problem. >>> >>>Hello Franz, >>> >>>I've started it on my Athlon 1500+ with all 3&4 piece TBs, and very few 5-piece >>>TBs. Depth 12 needs a minute and shows a factor around 3 between depths, >>>so I'm not sure this has any hope. >>> >>>Which 5-piece tables would be most useful here? >>>I could try to get some more... >>>[depth 13: 244 secs, factor 3.79... :-( ] >> >>Hello Heiner, >> >>I´m not absolutely sure, but I´d guess KNPKP would be the best one (since >>white can quite simply force it to remove both queens and win the black knight). > >I tried to download that one, but failed. Bob's site currently is out of order >and the European mirror I tried (ftp://145.94.41.27/Crafty/TB/32p/) >does provide not much more than 4KB/s, and after 17 MB the connection froze >completely. > >Just now I have: >FEN: 6k1/5Np1/4n3/2Qq2PK/8/8/8/8 w - - >analysing (mate in 20 moves): ># 1 0.00s 0kN 0.87 1- 0 ># 2 0.00s 0kN 1.00 1- 0 ># 3 0.00s 0kN [ 5.20] 0.92 30- 0 >% EGTB found tables for max 5 pieces >% EGTB uses 3728.4K memory internally ># 4 0.02s 0kN [ 3.67] 1.07 117- 0 ># 5 0.02s [ 1.00] 2kN [ 3.82] 1.51 317- 0 ># 6 0.07s [ 3.50] 11kN [ 6.22] 1.82 1444- 0 ># 7 0.23s [ 3.29] 51kN [ 4.51] 2.34 5591- 0 ># 8 0.94s [ 4.09] 243kN [ 4.76] 2.90 22069- 0 ># 9 3.07s [ 3.27] 726kN [ 2.98] 3.31 70315- 0 ># 10 6.83s [ 2.22] 1559kN [ 2.15] 4.02 155844- 1 ># 11 20.39s [ 2.99] 4502kN [ 2.89] 4.74 429776- 12 ># 12 64.43s [ 3.16] 13345kN [ 2.96] 6.58 1143927- 2685 ># 13 244.35s [ 3.79] 48544kN [ 3.64] 8.60 3750376- 1285851 ># 14 889.90s [ 3.64] 179343kN [ 3.69] 9.96 13432518- 10933117 ># 15 3422.06s [ 3.85] 670844kN [ 3.74] 12.51 47288122- 44788721 ># 16 14485.35s [ 4.23] 2862927kN [ 4.27] 12.36 193945070- 191445669 ># 17 59920.91s [ 4.14] 12006250kN [ 4.19] 12.45 801825148- 799325747 > >>Of course 6-men TBs would be still better, but probably they are not supported >>yet in your current version. (?) >>But IMO this is much more a position for a normal chess engine (and EGTBs) than >>for a mate solver, and (again only IMO) I don´t really see the need of EGTB- >>support in a mate solver at all - there are much too little positions which >>would profit by it, but OTH it will probably slowdown the search. >> >>What´s really surprising with this position, is that Chest is much slower than >>Gustav here (which is absolutely not usual!), and furthermore that the branching >>factor in Chest is significantly higher than that of Gustav (being about 2.5)!? > >Obviously, the branching factor causes the slowness (not vice versa). >And Chest's worse branching factor most probably is due to its selection >of defender moves. While Chest is not naive about this, it is far from >perfect, and sometimes a bit of "luck" is involved, also. > >For a more detailed explanation we had to compare the search trees of >Chest and Gustav in more detail. > >But when Gustav's branching factor here is as good as 2.5 there should >be a good chance to have Gustav complete depth 20, right? Have you tried? Hello Heiner, no, I didn´t try it - if you remember, I´m usually working on a slow Celeron/400 and it needed already 3 1/2 min to finish depth 13! So if I extrapolate the factor 2.5 for another 7 moves, this will take about 1 1/2 days - I doubt my more than 5 years old notebook will bear this torture! :-( Sorry that you had no luck with this KNPKP file, but since I´m quite sure that this position is indeed only a mate in 20, I´d guess that even this file won´t be of much use for a brute force search - if there really exists a shorter mate, then certainly not with only those 5 pieces. But I think this special position isn´t really interesting for mate solving, so we shouldn´t waste more of our time with it ... ;-) Regards, Franz.
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.