Author: Uri Blass
Date: 08:51:02 10/16/03
Go up one level in this thread
On October 16, 2003 at 10:48:52, Tord Romstad wrote: >On October 16, 2003 at 10:02:58, Uri Blass wrote: > >>On October 16, 2003 at 09:18:27, Tord Romstad wrote: >> >>>I would guess that most strong programs finds the right move rather quickly, >>>but that CM9K and Ruffian have one of their few blind spots in this particular >>>position. >> >>Your guess is wrong. > >This is really surprising to me, but OK. > >>most programs are different than your program and your program has some >>advantages relative to the top programs so the top programs cannot see the >>following line in a reasonable time: >> >>1. Nf6+ Kg7 2. Nh5+ Kg6 3. Bc2+ Kxh5 4. d8=Q Nf7+ 5. Ke6 Nxd8+ 6. Kf5 e2 7. >>Be4 e1=N 8. Bd5 c2 9. Bc4 c1=N 10. Bb5 Nc6 11. Bxc6 Nc7 12. Ba4 Nc2 13. >>Bxc2 Ne2 14. Bd1 c4 15. Bxe2# >>* >> >>They probably assume that black is winning because black has a big material >>advantage. >> >>I guess that your program extend moves that threat mate in 2 like >>10.Bb5 when other programs do not do it(even if they extend mate threat they may >>not do it in positions when the side to move has only a bishop and pawn because >>they were not designed to solve studies and in games it may be better not to >>waste time on checking it). > >I don't extend for mate threats in 2 moves or more, only for threats >of mate in 1 (and only in cases when I am lucky enough to detect the >mate threat in the null move search, which of course doesn't always >happen). However, threats of mate in 2 will sometimes be extended >as an effect of the Botvinnik-Markoff extension. > >As an experiment, i let my program analyse the same position with the >BM extension disabled. The result was rather interesting: The move >Nf6+ is still found at ply 13 I guess that your program find the line that I gave 1. Nf6+ Kg7 2. Nh5+ Kg6 3. Bc2+ Kxh5 4. d8=Q Nf7+ 5. Ke6 Nxd8+ 6. Kf5 e2 7. Be4 e1=N 8. Bd5 c2 9. Bc4 c1=N 10. Bb5 Nc6 11. Bxc6 Nc7 12. Ba4 Nc2 13. Bxc2 Ne2 14. Bd1 c4 15. Bxe2# let count plies(I assume that you extend every check and every mate threat by 1 ply and every push pawn to the 7th is extended by 1 ply and no pruning and still the problem is 19 plies) 1.Nf6+ Kg7 1 ply 2.Nh5+ Kg6 2 plies 3.Bc2+ Kxh5 3 plies 4.d8Q Nf7+ 4 plies 5.Ke6 Nxd8+ 5 plies 6.Kf5 e2 6 plies 7.Be4 threat mate e1N 7 plies 8.Bd5 c2 8 plies 9.Bc4 c1N 10 plies(Bc4 only threats mate in 2) 10.Bb5 Nc6 (12 plies Bb5 only threats mate in 2) 11.Bc6(threat mate) Nc7(13 plies) 12.Ba4 Nc2 (15 plies) 13.Bxc2 Ne2 (17 plies) if you call qsearch at this point you are not going to see the mate. 14.Bd1(18 plies) again qsearch is not going to help you to see the mate when it is black to move 14...c4(19 plies). only after 19 plies qsearch can return the mate. , but the program about 10% more nodes to >find it! This is very counter-intuitive to me, but I have observed it >in other positions as well. Sometimes Gothmog needs fewer nodes to >complete a search to a given depth with the BM extension than without. I am not so surprised because the extension may help it to reject illogical lines in previous plies so it has better information in the hash tables or in killer moves. Uri > >But still, the BM extension doesn't explain why my engine finds Nf6 >in a few minutes, as it was only slightly slower without the extension. >If it is really true that other programs have problems with this position, >I cannot find any explanation why my program is faster. Apart from the >Botvinnik-Markoff extension, I don't have any extensions except those >which almost everybody uses. > >Tord I guess that you have something that not almost everybody use. Uri
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.