Stockfish MateFinder Update

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Nordlandia
Posts: 2631
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Stockfish MateFinder Update

Post by Nordlandia » Wed Apr 13, 2016 6:22 pm

jhellis3 wrote:Certainly, if you have a version of SF which supports 6man nalimov TBs, you can use that :).
Already downloaded DTM 3-4-5 [7 GB]

Fritz 15 and Houdini supports Nalimov TBs.

What kind of performance can i expect in case i probe Nalimov 6-Men from USB 3.0 External HDD?

Of course 2 TB SSD will inflict better performance if you can afford it.

2 TB SSD can store 6-Men Nalimov + alot more personal stuff.
Last edited by Nordlandia on Wed Apr 13, 2016 6:28 pm, edited 1 time in total.

jhellis3
Posts: 459
Joined: Fri Aug 16, 2013 10:36 pm

Re: Stockfish MateFinder Update

Post by jhellis3 » Wed Apr 13, 2016 6:26 pm

What kind of performance can i expect in case i probe Nalimov 6-Men from USB 3.0 External HDD?
I have no earthly idea, I don't have Nalimov TBs on my system...

User avatar
Nordlandia
Posts: 2631
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Stockfish MateFinder Update

Post by Nordlandia » Wed Apr 13, 2016 6:45 pm

Nalimov performs better than syzygy in case using matefinder, that is atleast my observation.

The issue with syzygy is that even though stockfish recognize it as won position, it have to calculate DTM on its own.

4044: Barda=O - (+4046.12f5h8), Sjakkliv @2 1947
[d]b5qk/4Q1p1/5BP1/5K2/3p4/7n/7n/8 w - - 0 1

Analysis by Houdini 4 Pro x64:

1.Bxg7+ Qxg7 2.Qd8+ Qg8 3.Qxd4+ Qg7 4.Qh4+ Kg8 5.Qc4+ Bd5 6.Qxd5+ Kh8 7.Qd8+ Qg8 8.Qd4+ Qg7 9.Qh4+ Kg8 10.Qc4+ Kh8 11.Qc8+ Qg8 12.Qc3+ Qg7 13.Qxh3+ Kg8 14.Qb3+ Kh8 15.Qb8+ Qg8 16.Qxh2+
+- (#23) Depth: 34/58 00:00:06 178MN, tb=178149
(13.04.2016)

jhellis3
Posts: 459
Joined: Fri Aug 16, 2013 10:36 pm

Re: Stockfish MateFinder Update

Post by jhellis3 » Wed Apr 13, 2016 7:09 pm

1) It makes no sense to compare different engine outputs to measure the performance of TB implementations.

2) Obviously, TBs which support a DTM score are going to return an exact mate quicker than DTZ, all else being equal.

3) Neither SF nor SFMF support Nalimov TBs, and I do not have plans to do so. If you wish to discuss the efficacy of various TB implementations, please do so in another thread. This thread is, as the title suggest, for SFMF.


If you have a specific complaint or comment regarding SFMF, I am more than happy to answer it.

User avatar
Nordlandia
Posts: 2631
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Stockfish MateFinder Update

Post by Nordlandia » Wed Apr 13, 2016 7:29 pm

Alright, i'll probably have to wait until Houdini 5 (Houdini currently supports Nalimov format)

In which scenario shall i enable "Study" parameter?

Is there any example positions to visualize the difference between stock settings and in case study is enabled.

User avatar
MikeB
Posts: 4059
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Stockfish MateFinder Update

Post by MikeB » Thu Apr 14, 2016 2:26 am

Nordlandia wrote:Nalimov performs better than syzygy in case using matefinder, that is atleast my observation.

The issue with syzygy is that even though stockfish recognize it as won position, it have to calculate DTM on its own.

4044: Barda=O - (+4046.12f5h8), Sjakkliv @2 1947
[d]b5qk/4Q1p1/5BP1/5K2/3p4/7n/7n/8 w - - 0 1

Analysis by Houdini 4 Pro x64:

1.Bxg7+ Qxg7 2.Qd8+ Qg8 3.Qxd4+ Qg7 4.Qh4+ Kg8 5.Qc4+ Bd5 6.Qxd5+ Kh8 7.Qd8+ Qg8 8.Qd4+ Qg7 9.Qh4+ Kg8 10.Qc4+ Kh8 11.Qc8+ Qg8 12.Qc3+ Qg7 13.Qxh3+ Kg8 14.Qb3+ Kh8 15.Qb8+ Qg8 16.Qxh2+
+- (#23) Depth: 34/58 00:00:06 178MN, tb=178149
(13.04.2016)
Actually , in game play Nalimov is defective since it fails to understand the 50 move rule. That has been demostrated in numerous posts on this site.

Jouni
Posts: 2121
Joined: Wed Mar 08, 2006 7:15 pm

Re: Stockfish MateFinder Update

Post by Jouni » Thu Apr 14, 2016 3:00 pm

Really impressive! Latest version beats Houdini 3 tactical in many testsuites e.g. Brilliant set and also in mate solving. Of course I set "study" always when SFMF is not for playing :wink: .
But this old position [D]1q1r3k/3P1pp1/ppBR1n1p/4Q2P/P4P2/8/5PK1/8 w - - 0 1 is still difficult, when Houdini solves instantly.
Jouni

Jouni
Posts: 2121
Joined: Wed Mar 08, 2006 7:15 pm

Re: Stockfish MateFinder Update

Post by Jouni » Thu Apr 21, 2016 12:07 pm

Here one example showing matefinder name is well deserved. This is all 20 movers from Chest database (old version).

K4bk1/1r4p1/1qb2pnR/3p2NR/7r/7R/1p5n/QRRRRRRR w - - bm #20; 00:00;
1kr1n3/pp4nr/8/Q1PN1p2/6p1/5bPq/3R4/2BR2K1 w - - bm #20; 00:00 @ C3/R0/K6/P2/X5;
1r6/5q1r/1n4p1/2N2p2/p3p3/p1KpR1R1/b1B5/k7 w - - bm #20; 00:00 @ C0/R1/K2/P1/X2;
2n4k/2KB1ppp/1p3p2/p5Q1/4p1N1/Pbpr4/1Br3PR/1n1q4 w - - bm #20; 00:00 @ C2/R1/K3/P6/X25;
3K1B2/1p6/pp6/rk2N3/b1p5/1pP5/1P2P3/8 w - - bm #20; 00:00 @ C2/R0/K2/P1/X1;
3nn3/2p2p2/6p1/n1PB3n/3BK3/1P2N1P1/3pNPnk/3n3b w - - bm #20; 00:00 @ C3/R0/K2/P1/X1;
3nn3/2p2p2/6p1/n1PB3n/3BK3/1P2NRP1/3pNPnk/3n3b w - - bm #20; 00:00 @ C3/R0/K2/P1/X1;
4k3/2Q5/8/8/8/8/3p1pp1/3Kbrrb w - - bm #20; 00:00 @ C2/R0/K4/P1/X3;
4Q3/4p3/4b3/8/8/K7/p1r5/k7 w - - bm #20; 00:00 @ C0/R0/K3/P3/X8;
5K2/k7/1R4p1/1R1P4/3P4/5b2/5b2/5nq1 w - - bm #20; 00:00 @ C2/R1/K3/P0/X26;
6Qb/Pr5P/qP2b3/1R6/8/n6r/k6P/7K w - - bm #20; 00:00 @ C0/R1/K3/P5/X32;
7b/5p2/4pN1k/1p2P2P/1Pp5/K1p5/2P4B/8 w - - bm #20; 00:00 @ C2/R0/K2/P1/X1;
7k/prpp1K2/1P2p2q/7P/8/7p/3p3p/3Q4 w - - bm #20; 00:00 @ C0/R4/K3/P7/X31;
7Q/prpp1K1P/np6/8/8/5p2/pN3p2/k1b2r1b w - - bm #20; 00:00 @ C2/R1/K4/P9/X23;
8/1N6/p1R3p1/p2k1prb/p4Kpb/p5p1/P3P1P1/8 w - - bm #20; 00:00 @ C2/R0/K2/P1/X1;
8/1R1B4/K1p5/2p5/2p3p1/B7/p1p2p2/k7 w - - bm #20; 00:00 @ C2/R1/K2/P2/X8;
8/4p3/1K1pPp2/p2P1P2/BP2NP2/4p1p1/4pprb/4kbrn w - - bm #20; 00:00 @ C2/R0/K3/P1/X2;
8/4p3/3pPp2/pK1P1P2/1P2NP2/4p1p1/2B1pprb/4kbrn w - - bm #20; Duals(2); 00:00 @ C2/R0/K3/P1/X4;
8/8/8/8/1Q6/3k4/3p1pp1/3Kbrrb w - - bm #20; 00:00 @ C2/R0/K4/P1/X3;
8/K1p1pN2/n1B3p1/2kP1p2/P3p1P1/1pP5/1p1N1p2/1r6 w - - bm #20; 00:00 @ C2/R2/K2/P8/X23;
B2k4/KPpP4/n1Pb4/P5p1/5p2/5P2/8/8 b - - bm #20; 00:00 @ C2/R0/K3/P2/X2;
bbnK2Q1/8/1p1k4/r2P3q/P3P3/Pnp1p1p1/4PN2/8 w - - bm #20; 00:00 @ C3/R0/K3/P2/X2;
n2k4/1Kp2p2/2P2pp1/2B5/1p2R3/P2pp3/6p1/rb2bq2 w - - bm #20; 00:00 @ C0/R1/K3/P11/X31;
nB6/2p1p3/1p2Pp2/1P2kp2/2Kp1N2/1PpP1pP1/2P2P2/8 w - - bm #20; 00:00 @ C2/R0/K2/P2/X3;
Q7/6p1/bp4p1/kp6/8/1P6/6p1/6Kb w - - bm #20; 00:00 @ C2/R0/K2/P4/X11;
qb6/n6P/3B4/p7/8/p7/p7/k2K4 w - - bm #20; 00:00 @ C2/R1/K3/P2/X3;
r7/4r1P1/3Knb2/1RNR3B/5k1q/3P4/2b3P1/5N2 w - - bm #20; 00:00 @ C0/R4/K2/P1/X1;
rR4r1/6P1/8/8/p5Q1/k7/1p2n2K/qb6 w - - bm #20; 00:00 @ C3/R0/K2/P2/X2;
8/8/6k1/Q7/8/8/3p1pp1/3Kbrrb w - - bm #20; 00:00+ @ C2/R0/K6/P1/X5;
8/6p1/p7/rp1K2p1/kb3pP1/2p2p1b/P1Np1P2/3N2R1 w - - bm #20; 00:01 @ C0/R0/K1/P3/X9;
bqB2n1r/1P2p2p/1p2p3/2P1RP1k/4Pr1P/1P3n2/Kp3Np1/b1B2N1R w - - bm #20; 00:01 @ C0/R1/K4/P9/X46;
8/8/2k5/8/3p4/Qp1p4/pp1p4/rrbK4 w - - bm #20; 00:02 @ C2/R0/K7/P1/X6;
2B4k/1r2B1p1/1P2p3/pN1bn3/R7/8/2Q1R1p1/K2n1q2 w - - bm #20; 00:14 @ C2/R1/K5/P7/X39;
4K3/p5p1/p5p1/P1p3P1/B1P5/2P4p/6np/3R2bk w - - bm #20; 00:18+ @ C0/R0/K1/P1/X9;
1K4kB/6P1/6P1/3r4/8/1R6/B1PP4/7R w - - bm #20; 02:16 @ C0/R0/K3/P0/X15;
8/2p4Q/8/8/p7/K6p/2r5/1k6 w - - bm #20; --:-- @ C0/R16/K7/P4/X15;
8/p5p1/4p3/4Pp2/2R2P1K/3P1P2/pk1p4/qrbB4 w - - bm #20; --:-- @ C0/R0/K3/P3/X5;
8/RB6/8/8/8/5bP1/4Nprp/5Kbk w - - bm #20; --:-- @ C0/R0/K1/P2/X13;
Q1B5/P2Pn1p1/1rNpb3/2kn2p1/8/1P3P2/1NpBKPPp/b6q w - - bm #20; --:-- @ C0/R3/K5/P9/X41;
qrb5/rk1p4/p2P4/Pp6/1N2n3/6p1/5nBb/5K2 w - - bm #20; --:-- @ C0/R-12/K2/P3/X20; 00'29" (Gustav);
RN2r2k/6pP/6P1/5K2/8/8/6R1/8 w - - bm #20; --:-- @ C0/R-12/K4/P2/X14;
1rB2r1B/3pp1N1/3qp1n1/QRppp1K1/3b1Pp1/3bp3/4NPk1/1n1R4 w - - bm #20;
5K1k/1R1P4/8/b7/4r3/3B4/3p4/8 w - - bm #20;
7n/7P/8/8/8/7Q/5ppR/K4bkn w - - bm #20;
8/8/3p4/N2p4/1k6/p2K4/8/1B6 w - - bm #20;
qrr1b3/pn2p3/npppp1P1/4pP2/2P2BPp/PPN1P2B/4R2K/2k5 w - - bm #20;

I gave Houdini3 and SFMF one minute and 4 cores to solve. SF in study mode Houdini default (mate mode isn't better). Solved positions

Code: Select all

 1 SH
 2 SH
 3 SH
 4 SH
 5  H
 6 SH
 7 SH
 8 
 9 SH
10 SH
11 SH
12
13 SH
14 SH
15 S
16 SH
17 S
18 S
19  H
20 S
21 SH
22 SH
23 SH
24 
25 SH
26 SH
27 SH
28 SH
29 
30
31 SH
32 S
33 SH
34 
35 SH
36
37
38
39 SH
40 S
41 S
42  H
43
44 S
45
46
Stockfish solved 31 and Houdini 26. BTW position 15 is actually mate in 17 as SF says!

[D]8/1N6/p1R3p1/p2k1prb/p4Kpb/p5p1/P3P1P1/8 w - - 0 1

Analysis by SF MF110416 64 bmi2:

+- (#17) Depth: 46/56 00:00:27 240mN, tb=36054
1.Rd6+ Kc4 2.Rd3 Kb4 3.Nd6 Kc5 4.Nf7 Kc6 5.Ne5+ Kc7 6.Rd1 Kc8 7.Rd7 Kb8 8.Nf7 Kc8 9.Re7 Kb8 10.Nd6 Ka8 11.Rc7 Kb8 12.Rb7+ Ka8 13.Rb1 Ka7 14.Nc4 Ka8 15.Nxa5 Ka7 16.Nc6+ Ka8 17.Rb8#
Jouni

User avatar
F.Huber
Posts: 794
Joined: Thu Mar 09, 2006 3:50 pm
Location: Austria
Contact:

Re: Stockfish MateFinder Update

Post by F.Huber » Thu Apr 21, 2016 2:38 pm

Jouni wrote: Stockfish solved 31 and Houdini 26. BTW position 15 is actually mate in 17 as SF says!
Yep, and ChestUCI confirms this (it even has 2 solutions):
8/1N6/p1R3p1/p2k1prb/p4Kpb/p5p1/P3P1P1/8 w - - bm #17; Duals(2); 00:44 @ C6/R0/K6/P1/X5;

Once again just a problem of how much you restrict the search ...

Franz

User avatar
Eelco de Groot
Posts: 4245
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Stockfish MateFinder Update

Post by Eelco de Groot » Thu Apr 21, 2016 3:35 pm

Rainbow Serpent can find the second solution but it takes a bit longer to see it is Mate in 17
8/1N6/p1R3p1/p2k1prb/p4Kpb/p5p1/P3P1P1/8 w - -

Engine: Rainbow Serpent 20162803_016 MP (i7 6700, 4 threads, 512 MB)
by T. Romstad, M. Costalba, J. Kiiski, G. Linscott
.
.
.
47/65 3:23 +M20 1.Nd8 Kd4 2.Nf7 Kd5 3.Ne5 Kd4 4.Rc1 Kd5
5.Rd1+ Ke6 6.Rd7 Kf6 7.Rd6+ Ke7
8.Rb6 Kd8 9.Rb7 Kc8 10.Rd7 Kb8
11.Re7 Kc8 12.Nf7 Kb8 13.Nd6 Ka8
14.Rd7 (1.639.460.072) 8062

48/65 3:57 +M19 1.Nd8 Kd4 2.Nf7 Kd5 3.Ne5 Kd4 4.Rd6+ Kc3
5.Rd8 Kc2 6.Nc4 Kc3 7.Ne3 Kb4 8.Rc8 Kb5
9.Rc7 Kb6 10.Nd5+ Kb5 11.Nf6 Kb6
12.Ne8 Kb5 13.Nd6+ Kb6 14.Rc1 (1.897.371.917) 7977

49/65 5:25 +M19 1.Nd8 Kd4 2.Nf7 Kd5 3.Ne5 Kd4 4.Rd6+ Kc3
5.Rd8 Kc2 6.Nc4 Kc3 7.Ne3 Kb4 8.Rc8 Kb5
9.Rc7 Kb6 10.Nd5+ Kb5 11.Nf6 Kb6
12.Ne8 Kb5 13.Nd6+ Kb6 14.Rc1 (2.557.058.387) 7862

50/65 6:55 +M19 1.Nd8 Kd4 2.Nf7 Kd5 3.Ne5 Kd4 4.Rd6+ Kc3
5.Rd8 Kc2 6.Nc4 Kc3 7.Ne3 Kb4 8.Rc8 Kb5
9.Rc7 Kb6 10.Nd5+ Kb5 11.Nf6 Kb6
12.Ne8 Kb5 13.Nd6+ Kb6 14.Rc1 (3.250.978.107) 7827

51/65 9:42 +M19 1.Nd8 Kd4 2.Nf7 Kd5 3.Ne5 Kd4 4.Rd6+ Kc3
5.Rd8 Kc2 6.Nc4 Kc3 7.Ne3 Kb4 8.Rc8 Kb5
9.Rc7 Kb6 10.Nd5+ Kb5 11.Nf6 Kb6
12.Ne8 Kb5 13.Nd6+ Kb6 14.Rc1 (4.513.097.906) 7754

52/65 12:02 +M19 1.Nd8 Kd4 2.Nf7 Kd5 3.Ne5 Kd4 4.Rd6+ Kc3
5.Rd8 Kc2 6.Nc4 Kc3 7.Ne3 Kb4 8.Rc8 Kb5
9.Rc7 Kb6 10.Nd5+ Kb5 11.Nf6 Kb6
12.Ne8 Kb5 13.Nd6+ Kb6 14.Rc1 (5.587.711.526) 7732

53/65 17:39 +M19 1.Nd8 Kd4 2.Nf7 Kd5 3.Ne5 Kd4 4.Rd6+ Kc3
5.Rd8 Kc2 6.Nc4 Kc3 7.Ne3 Kb4 8.Rc8 Kb5
9.Rc7 Kb6 10.Nd5+ Kb5 11.Nf6 Kb6
12.Ne8 Kb5 13.Nd6+ Kb6 14.Rc1 (8.142.465.466) 7688

54/65 24:28 +M19 1.Nd8 Kd4 2.Nf7 Kd5 3.Ne5 Kd4 4.Rd6+ Kc3
5.Rd8 Kc2 6.Nc4 Kc3 7.Ne3 Kb4 8.Rc8 Kb5
9.Rc7 Kb6 10.Nd5+ Kb5 11.Nf6 Kb6
12.Ne8 Kb5 13.Nd6+ Kb6 14.Rc1 (11.184.819.885) 7614

55/65 32:12 +M17 1.Nd8 Kd4 2.Nf7 Kd5 3.Rd6+ Kc5 4.Rd3 Kc4
5.Ne5+ Kb4 6.e3 Kc5 7.Rd2 Kb4 8.Rc2 Kb5
9.Nf7 Kb4 10.Rc7 Kb5 11.Nd6+ Kb6
12.Rc1 Ka7 13.Rb1 Ka8 14.Nc4 (14.679.922.043) 7597


best move: Nb7-d8 time: 32:59.203 min n/s: 7.597.611 nodes: 15.032.880.677
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan

Post Reply