Page 1 of 5
MateFinder 9
Posted: Mon Feb 12, 2018 6:57 pm
by jhellis3
I have released MateFinder 9 here:
https://github.com/jhellis3/Stockfish/releases/tag/MF_9
Bench is 6898833.
It includes the recently merged dynamic contempt, however, I have made this user selectable via UCI checkbox and the default is set to off.
Additionally, the default contempt is set to 0. If one wants to emulate current SF behavior, they will need to set contempt to 18, and check "Dynamic Contempt" in the UCI settings.
I have also provided some very basic builds. However, please note these will not be fast at all. If you want to use MateFinder for competitive purposes I highly recommend compiling it yourself.
I will also be happy to host high-performance builds should anyone desire to provide them.
Re: MateFinder 9
Posted: Tue Feb 13, 2018 5:04 am
by tmokonen
Thank you, Joseph.
Re: MateFinder 9
Posted: Tue Feb 13, 2018 6:02 am
by Vinvin
tmokonen wrote:Thank you, Joseph.
+1
Re: MateFinder 9
Posted: Tue Feb 13, 2018 10:12 am
by Dann Corbit
Just tried it on a collection of mates and it did quite well.
Re: MateFinder 9
Posted: Tue Feb 13, 2018 7:38 pm
by Jouni
When testing mates from Chest database I found slightly surprising thing. Matefinder (2017) was clearly best, when solving mates for 11-20 movers. When solving 6-10 movers SF9 and Houdini6(tactical!) were equal or even faster. I have no clue why? But of course 6-10 movers should be solved all even with limited time? I will test new matefinder later, when there is fast compile.
Re: MateFinder 9
Posted: Thu Feb 15, 2018 4:36 pm
by zullil
Dann Corbit wrote:Just tried it on a collection of mates and it did quite well.
[D] rk1b4/p2p2p1/1P6/2R2P2/8/2K5/8/5B2 w - - 0 1
MateFinder is struggling with this one! But it's quite a hard problem. Maybe Sting or McBrain will prove more successful.
Code: Select all
info depth 52 seldepth 55 multipv 1 score cp -8 nodes 35647568533 nps 33955110 hashfull 1000 tbhits 0 time 1049844 pv c5d5 d8f6 c3b4 a7b6 d5d7 a8a7 d7a7 b8a7 f1b5 a7b7 b4c4 b7c7 c4d5 f6c3 d5e6 c7d8 b5a4 d8c8 a4b5 c8c7 b5d7 c3f6 d7b5 f6b2 e6d5 b2a3 d5e6 a3c5 b5d7 c5b4 d7b5 b4a3 e6d5 a3c5 d5e6 c5e3 e6e5 e3g5 e5d5 g5f4 d5e6 c7b7 e6d5 f4e3 d5c4 e3d2 c4d4 b7c7 d4d5 d2g5 d5e5 g5e7 e5d5 e7f6
Re: MateFinder 9
Posted: Thu Feb 15, 2018 8:51 pm
by jhellis3
IpMan has created some higher performance complies, and I have updated the release page to include them.
https://github.com/jhellis3/Stockfish/releases/tag/MF_9
Alternatively, you may find them at his site:
http://www.ipmanchess.yolasite.com/compiles.php
Re: MateFinder 9
Posted: Fri Feb 16, 2018 1:30 pm
by zullil
zullil wrote:Dann Corbit wrote:Just tried it on a collection of mates and it did quite well.
[D] rk1b4/p2p2p1/1P6/2R2P2/8/2K5/8/5B2 w - - 0 1
MateFinder is struggling with this one! But it's quite a hard problem. Maybe Sting or McBrain will prove more successful.
Code: Select all
info depth 52 seldepth 55 multipv 1 score cp -8 nodes 35647568533 nps 33955110 hashfull 1000 tbhits 0 time 1049844 pv c5d5 d8f6 c3b4 a7b6 d5d7 a8a7 d7a7 b8a7 f1b5 a7b7 b4c4 b7c7 c4d5 f6c3 d5e6 c7d8 b5a4 d8c8 a4b5 c8c7 b5d7 c3f6 d7b5 f6b2 e6d5 b2a3 d5e6 a3c5 b5d7 c5b4 d7b5 b4a3 e6d5 a3c5 d5e6 c5e3 e6e5 e3g5 e5d5 g5f4 d5e6 c7b7 e6d5 f4e3 d5c4 e3d2 c4d4 b7c7 d4d5 d2g5 d5e5 g5e7 e5d5 e7f6
Terminating search. Been "stuck" at depth 53 for almost 24 hours now.
Code: Select all
stop
info depth 53 seldepth 55 multipv 1 score cp -8 nodes 2777671982916 nps 35012515 hashfull 1000 tbhits 0 time 79333688 pv c5d5 d8f6 c3b4 a7b6 d5d7 a8a7 d7a7 b8a7 f1b5 a7b7 b4c4 b7c7 c4d5 f6c3 d5e6 c7d8 b5a4 d8c8 a4b5 c8c7 b5d7 c3f6 d7b5 f6b2 e6d5 b2a3 d5e6 a3c5 b5d7 c5b4 d7b5 b4a3 e6d5 a3c5 d5e6 c5e3 e6e5 e3g5 e5d5 g5f4 d5e6 c7b7 e6d5 f4e3 d5c4 e3d2 c4d4 b7c7 d4d5 d2g5 d5e5 g5e7 e5d5 e7f6
bestmove c5d5 ponder d8f6
Re: MateFinder 9
Posted: Sat Feb 17, 2018 4:22 pm
by lech
zullil wrote:
[D] rk1b4/p2p2p1/1P6/2R2P2/8/2K5/8/5B2 w - - 0 1
MateFinder is struggling with this one! But it's quite a hard problem. Maybe Sting or McBrain will prove more successful.
Sting SF 9.8 returns move Rc8+ immediately. After 1 minutes (old dual core) gives a solid gain too.
Code: Select all
info depth 33 multipv 1 score cp 0 nodes 45870180 nps 848661 time 54050 pv c5c8b8c8 f1a6 c8b8 b6b7 d8f6 c3c4
info depth 34
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 8 lowerbound nodes 58422165 nps 858770 time 68030 pv c5c8
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 16 lowerbound nodes 60533834 nps 860112 time 70379 pv c5c8
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 28 lowerbound nodes 65900683 nps 862574 time 76400 pv c5c8 b8c8 f1a6 c8b8 b6b7 d8f6 c3c4 f6e5 c4c5 e5f4 c5d5 f4c7 d5e4 c7d8 e4d3 d8e7 d3c4 e7f8 c4d5 f8e7 d5e5 e7f8 f5f6 g7f6 e5f6 f8c5 f6f7 c5e3 f7e7 d7d5 e7d7 d5d4 b7a8q b8a8 d7c8 d4d3 a6b7
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 46 lowerbound nodes 68055435 nps 864580 time 78
715 pv c5c8 b8c8 f1a6 c8b8 b6b7 d8f6 c3c4 f6e5 c4c5 e5f4 c5d5 f4c7 d5e4 c7d8 e4d3 d8e7 d3c4 e7f8 c4d5 f8e7 d5e5 e7f8 f5f6 g7f6 e5f6 f8c5 f6f7 c5e3 f7e7 d7d5 e7d7 d5d4 b7a8q b8a8 d7c8 d4d3 a6b7
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 73 lowerbound nodes 71677974 nps 867163 time 82658 pv c5c8 b8c8 f1a6 c8b8 b6b7 d8f6 c3c4 f6e5 c4c5 e5f4 c5d5 f4c7 d5e4 c7d8 e4d3 d8e7 d3c4 e7f8 c4d5 f8e7 d5e5 e7f8 f5f6 g7f6 e5f6 f8c5 f6f7 c5e3 f7e7 d7d5 e7d7 d5d4 b7a8q b8a8 d7c8 d4d3 a6b7
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 114 lowerbound nodes 74160406 nps 869223 time 8
5318 pv c5c8 b8c8 f1a6 c8b8 b6b7 d8f6 c3c4 f6e5 c4c5 e5f4 c5d5 f4c7 d5e4 c7d8 e4d3 d8e7 d3c4 e7f8 c4d5 f8e7 d5e5 e7f8 f5f6 g7f6 e5f6 f8c5 f6f7 c5e3 f7e7 d7d5 e7d7 d5d4 b7a8q b8a8 d7c8 d4d3 a6b7
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 175 lowerbound nodes 76304696 nps 869301 time 87777 pv c5c8 b8c8 f1a6 c8b8 b6b7 d8f6 c3c4 f6e5 c4c5 e5f4 c5d5 f4c7 d5e4 c7d8 e4d3 d8e7 d3c4 e7f8 c4d5 f8e7 d5e5 e7f8 f5f6 g7f6 e5f6 f8c5 f6f7 c5e3 f7e7 d7d5 e7d7 d5d4 b7a8q b8a8 d7c8 d4d3 a6b7
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 267 lowerbound nodes 77824266 nps 869642 time 8
9490 pv c5c8 b8c8 f1a6 c8b8 b6b7 d8f6 c3c4 f6e5 c4c5 e5f4 c5d5 f4c7 d5e4 c7d8 e4d3 d8e7 d3c4 e7f8 c4d5 f8e7 d5e5 e7f8 f5f6 g7f6 e5f6 f8c5 f6f7 c5e3 f7e7 d7d5 e7d7 d5d4 b7a8q b8a8 d7c8 d4d3 a6b7
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 404 lowerbound nodes 84539562 nps 866148 time 97604 pv c5c8 b8c8 f1a6 c8b8 b6b7 d8f6 c3c4 f6e5 c4c5 e5f4 c5d5 f4c7 d5e4 c7d8 e4d3 d8e7 d3c4 e7f8 c4d5 f8e7 d5e5 e7f8 f5f6 g7f6 e5f6 f8c5 f6f7 c5e3 f7e7 d7d5 e7d7 d5d4 b7a8q b8a8 d7c8 d4d3 a6b7
info currmove c5c8 currmovenumber 1
info depth 34 multipv 1 score cp 609 lowerbound nodes 95483271 nps 854987 time 1
11678 pv c5c8 b8c8 f1a6 c8b8 b6b7 d8f6 c3c4 f6e5 c4c5 e5f4 c5d5 f4c7 d5e4 c7d8 e4d3 d8e7 d3c4 e7f8 c4d5 f8e7 d5e5 e7f8 f5f6 g7f6 e5f6 f8c5 f6f7 c5e3 f7e7 d7d5 e7d7 d5d4 b7a8q b8a8 d7c8 d4d3 a6b7
info currmove c5c8 currmovenumber 1
info currmove c5d5 currmovenumber 2
info currmove c3d4 currmovenumber 3
info currmove c3d2 currmovenumber 4
info currmove c3b4 currmovenumber 5
info currmove c3b3 currmovenumber 6
info currmove c3d3 currmovenumber 7
info currmove c3c2 currmovenumber 8
info currmove c3b2 currmovenumber 9
info currmove c3c4 currmovenumber 10
info currmove c5c6 currmovenumber 11
info currmove b6a7 currmovenumber 12
info currmove b6b7 currmovenumber 13
info currmove f1a6 currmovenumber 14
info currmove c5a5 currmovenumber 15
info currmove f5f6 currmovenumber 16
info currmove f1b5 currmovenumber 17
info currmove f1g2 currmovenumber 18
info currmove c5e5 currmovenumber 19
info currmove c5c4 currmovenumber 20
info currmove c5b5 currmovenumber 21
info currmove f1d3 currmovenumber 22
info currmove f1e2 currmovenumber 23
info currmove f1c4 currmovenumber 24
info currmove f1h3 currmovenumber 25
info currmove c5c7 currmovenumber 26
info depth 34 multipv 1 score cp 650 nodes 113903295 nps 844954 time 134804 pv c5c8 b8b7 f1g2 b7b6 g2a8 d8f6 c3c4 a7a5 c8g8 a5a4 g8b8 b6c7 b8b7 c7d6 b7a7 d6e5 a7a5 e5f4 a8b7 a4a3 a5a3 f4f5 c4d5 f5f4 b7c8 d7d6 d5d6 f4e4 c8b7 e4d4
Re: MateFinder 9
Posted: Sat Feb 17, 2018 4:47 pm
by CMCanavessi
Shredder 13 also likes Rd5 and doesn't see Rc8+
Code: Select all
FEN: rk1b4/p2p2p1/1P6/2R2P2/8/2K5/8/5B2 w - - 0 1
Shredder 13 x64:
39/62 04:27,424 528.357k 1.976k -0,13 1.Rd5 Kb7 2.Rd6 Rc8+ 3.Kb3 Bxb6 4.Rxd7+ Rc7 5.Rd6 Rf7 6.Bc4 Rxf5 7.Rd7+ Kc6 8.Rxg7 Rf3+ 9.Ka4 Rf4 10.Kb3 a5 11.Rg6+ Kc7 12.Rg7+ Kb8 13.Rf7 Rd4 14.Rf5 Kb7 15.Bd5+ Kc7 16.Bc4 Kc6 17.Rf6+ Kb7 18.Kc3 Rh4 19.Bd5+ Ka6 20.Bf7 Ka7 21.Rc6 Bd8 22.Bc4 Rg4 23.Kb3 Kb7 24.Rc5 a4+ 25.Kxa4 Bb6 26.Kb4
Houdini 6.03 doesn't see it either
Code: Select all
FEN: rk1b4/p2p2p1/1P6/2R2P2/8/2K5/8/5B2 w - - 0 1
Houdini 6.03 x64:
36/65 04:36,805 623.710k 2.253k -0,25 1.Rd5 Bf6+ 2.Kb4 d6 3.Kb5 Be5 4.Bc4 axb6 5.Kxb6 Ra4 6.Bb3 Rf4 7.Kc6 Kc8 8.Rb5 Kd8 9.f6 gxf6 10.Rb8+ Ke7 11.Rb7+ Kf8 12.Kd5 Rf2 13.Ke6 Kg8 14.Rf7 Rg2 15.Bd5 Rg5 16.Bc4 Rg7 17.Rxf6 Bxf6 18.Kxf6+ Kh7 19.Ke6 Rg6+ 20.Kd7 Kh6 21.Be6 Kg5 22.Kxd6 Kf4 23.Kd5 Rg5+ 24.Kd6 Ke3 25.Bd5 Kd4 26.Be6 Ra5 27.Bd7