Sting SF 19

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.
lech
Posts: 1060
Joined: Sun Feb 14, 2010 9:02 pm

Sting SF 19

Post by lech » Fri Nov 29, 2019 9:26 am

download: http://www.mediafire.com/file/v0esphvbg ... 9.zip/file

I hope this version can be the last one. :)
Maybe, I can't be friendly, but let me be useful.

Paloma
Posts: 855
Joined: Thu Dec 25, 2008 8:07 pm

Re: Sting SF 19

Post by Paloma » Fri Nov 29, 2019 8:48 pm

Thanks Marek, what is new on Sting SF 19?

User avatar
Pafifi
Posts: 3
Joined: Sat Nov 16, 2019 1:43 am
Full name: Rong Lin

Re: Sting SF 19

Post by Pafifi » Sat Nov 30, 2019 2:33 am

I doubt if Sting SF is really getting stronger on solving puzzles.
n1QBq1k1/5p1p/5KP1/p7/8/8/8/8 w - - 0 1
For this puzzle version 15 solved it in a few seconds, while this version spent minutes.

lech
Posts: 1060
Joined: Sun Feb 14, 2010 9:02 pm

Re: Sting SF 19

Post by lech » Sat Nov 30, 2019 10:05 am

From the version 16, I started a serious reorganization of the Sting/s code.
Why?
At first, I removed (cleaned) all what could be seen as selective codes.
For this reason some positions can be solving in longer time than earlier.
But the main problem of solving many positions (long solutions) is the horizon effect.
Sting is able to analyze only 100 half-moves in chain of moves of alpha-beta searching.
It has no sense (a wrong idea) to increase it.

It seems to be better to skip good found moves (go forward) and to start searching or "auto-playing" from the new place to get a further horizon.
Sometimes, another good method can be to stop searching and to start again without clearing of hash.
And from the version 16 it was the main idea of my work to support it.

Be active when you try to use Sting!

What is new in version 19?
It is a first version which seems to satisfy me and can be the last one. :)
Maybe, I can't be friendly, but let me be useful.

lech
Posts: 1060
Joined: Sun Feb 14, 2010 9:02 pm

Re: Sting SF 19

Post by lech » Sat Nov 30, 2019 11:39 am

(!) I replaced the last item by a missing "perpetual check" example.

The "IQ" test is a very important part of my project (Sting) and doesn't contain middle-game tactical positions and technical endgames.
Here are positions dedicated to all engines to test so-called by me "quasi-intelligent" theirs abilities.
Set parameters: "threads" to 1 and "hash" to 1024 and clear hash before each of items.
Don't use such external tricks like: rule-50 changing, null-move disabling, special parameters and others.
The external tools like: MultiPV, tablebases, "backward-forward" and "Monte Carlo" are not welcome too.

"WIN" means the correct solution and some non-zero plus score.
Don't do the test in a machine way. Test and analyze each of items separately!



The authors of studies selected to the "IQ" test:

by YACPDB:

Blathy Otto - 3, 12, 13
Benko Pal - 16, 75
Hasek Josef - 22, 28
Gurvich Abram - 23
Kasparyan Genrikh - 24
Neghina Mihai - 27
Grasemann Herbert - 37
Popov Georgy & Hanyan Aleksei - 39 (part of study)
Rudolph William - 46
Troitzky Aleksei - 57
Reti Richard - 58
Matous Mario - 60
Behting Carl - 62
Klyatskin Mikhail - 66
Kubbel Karl - 69
Novomesky Daniel - 73
Lazard Frederic - 74
Marwitz Jan - 79
Lamford Paul - 82
Simkhovich Froim - 86

by other sources:

Solovyov Y. - 6
Babic M. - 9
Neghina Mihai - 11, 31, 84, 87, 89, 92
Simkhovich Froim - 20
van Breukelen Gijs - 36
Hoch Yehuda & Aloni Hillel - 55
Troitzky Aleksei - 59
Smyslov Vasily - 67
Bondarenko F. & Kuznetsov A. - 68
Surya Sekhar Ganguly - 81
Stavrietsky Aleksandr - 88
Kazantsev A. - 93
Ryabinin Nikolaj - 96
Neghina Mihai & Rusz Arpad- 97
Kamody Mario & Neghina Mihai - 98
Fekete Zoltan - 99 (part of study)
Maybe, I can't be friendly, but let me be useful.

User avatar
Pafifi
Posts: 3
Joined: Sat Nov 16, 2019 1:43 am
Full name: Rong Lin

Re: Sting SF 19

Post by Pafifi » Sun Dec 01, 2019 2:21 am

Another issue is the resource usage. I run it on my Windows Server 2008, 24 cores, 160G RAM. Even I set hash to 131072, the taskmgr said Sting only allocated 8G RAM, and CPU does not keep on 100%, unlike stockfish

lech
Posts: 1060
Joined: Sun Feb 14, 2010 9:02 pm

Re: Sting SF 19

Post by lech » Sun Dec 01, 2019 1:35 pm

Pafifi wrote:
Sun Dec 01, 2019 2:21 am
Another issue is the resource usage. I run it on my Windows Server 2008, 24 cores, 160G RAM. Even I set hash to 131072, the taskmgr said Sting only allocated 8G RAM, and CPU does not keep on 100%, unlike stockfish
Sting is based on the old version (2.1.1) of Stockfish.
Indeed, Sting can use "only" 8G (max) of Hash but I don't know: is much bigger (than 1024M) hash able to help something, or not?
Maybe, I can't be friendly, but let me be useful.

MikeB
Posts: 3539
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Sting SF 19

Post by MikeB » Sun Dec 01, 2019 10:47 pm

Pafifi wrote:
Sat Nov 30, 2019 2:33 am
I doubt if Sting SF is really getting stronger on solving puzzles.
n1QBq1k1/5p1p/5KP1/p7/8/8/8/8 w - - 0 1
For this puzzle version 15 solved it in a few seconds, while this version spent minutes.

Problem solving seems to come and go with any engine that I work on:
Black-Diamond XR7

Code: Select all

 19	  #12 	4.67M  	0:00.46	Bc7 Qxc8 gxf7+ Kh8 Be5 Qc5 Bb2 Nc7 Ba1 a4 Bb2 a3 Ba1 a2 Bb2 a1=Q Bxa1 Qc3+ Bxc3 Nd5+ Ke6+ Nxc3 f8=Q# 
 18	  #12 	4.62M  	0:00.46	Bc7 Qxc8 gxf7+ Kh8 Be5 Qc5 Bb2 Nc7 Ba1 a4 Bb2 a3 Ba1 a2 Bb2 a1=Q Bxa1 Qc3+ Bxc3 Nd5+ Ke6+ Nxc3 f8=Q# 
 17	  0.00 	2.69M  	0:00.29	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qd1 f6+ Kh4 Qe4+ Kg3 a4 Bb6 a3 Qd8 Qe5+ Kg4 Qe4+ 
 16	  0.00 	2.03M  	0:00.23	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qd1 f6+ Kh4 Qe4+ Kg3 a4 Bb6 a3 Qd8 Qe5+ Kg4 Qe4+ 
 15	  0.00 	1.66M  	0:00.20	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qd1 f6+ Kh4 Qe4+ Kg3 a4 Bb6 a3 Qd8 Qe5+ Kg4 Qe4+ 
 14	  0.00 	1.28M  	0:00.16	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qd1 f6+ Kh4 Qe4+ Kg3 a4 Bb6 a3 Qd8 Qe5+ Kg4 Qe4+ 
 13	  0.00 	1.08M  	0:00.14	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qd1 f6+ Kh4 Qe4+ Kg3 a4 Bb6 a3 Qd8 Qe5+ Kg4 Qe4+ 
 12	  0.00 	834033	0:00.12	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qd1 f6+ Kh4 Qe4+ Kg3 a4 Qd2 a3 Bb6 a2 Qxa2 
 11	  0.00 	659607	0:00.10	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qd1 f6+ Kh4 Qe4+ Kg3 a4 Qd2 a3 Bb6 a2 Qxa2 
 10	  0.00 	572752	0:00.09	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qd1 f6+ Kh4 Qe4+ Kg3 a4 Qd2 a3 Bb6 a2 Qxa2 
  9	  0.00 	530796	0:00.08	gxh7+ Kh8 Qc3 Qe6+ Kg5+ Kxh7 Qf6 Qxf6+ Bxf6 
  8	+0.19 	341281	0:00.07	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qd1 Qe5+ Kg4 a4 Bg5 Kg6 
  7	+0.25 	233350	0:00.06	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Kf4 Qd6+ Ke3 Qa3+ Ke4 Qa2 
  6	+0.47 	107031	0:00.05	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Kxh7 Qh1+ Kg7 Qf3 
  5	+0.24 	21889  	0:00.02	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Qe5+ Kg4 Kxh7 
  4	+0.35 	15219  	0:00.02	gxh7+ Kh8 Qxa8 Qe6+ Kg5 Qe5+ Kh6 
  3	+0.30 	7455    	0:00.00	gxh7+ Kh8 Qxa8 Qe6+ Kg5 
  2	+0.30 	5274    	0:00.00	gxh7+ Kh8 Qxa8 
  1	+0.32 	2383    	0:00.00	gxh7+ Kxh7 Qh3+ Kg8 
  0	# 
Sting is a great effort by Marek and one I appreciate.

User avatar
Pafifi
Posts: 3
Joined: Sat Nov 16, 2019 1:43 am
Full name: Rong Lin

Re: Sting SF 19

Post by Pafifi » Tue Dec 03, 2019 12:27 am

lech wrote:
Sun Dec 01, 2019 1:35 pm
Pafifi wrote:
Sun Dec 01, 2019 2:21 am
Another issue is the resource usage. I run it on my Windows Server 2008, 24 cores, 160G RAM. Even I set hash to 131072, the taskmgr said Sting only allocated 8G RAM, and CPU does not keep on 100%, unlike stockfish
Sting is based on the old version (2.1.1) of Stockfish.
Indeed, Sting can use "only" 8G (max) of Hash but I don't know: is much bigger (than 1024M) hash able to help something, or not?
It really helps when you need deep analysis. Hash stores evaluated positions. When the hash is nearly full after a long-time analysis, old positions will be replaced by new ones. However, the next time the replaced position is visited, the engine will have to re-evaluate it. That is to say, the larger the hash is, the less repetitive work the engine will do.

lech
Posts: 1060
Joined: Sun Feb 14, 2010 9:02 pm

Re: Sting SF 19

Post by lech » Tue Dec 03, 2019 9:03 am

Pafifi wrote:
Tue Dec 03, 2019 12:27 am
lech wrote:
Sun Dec 01, 2019 1:35 pm
Pafifi wrote:
Sun Dec 01, 2019 2:21 am
Another issue is the resource usage. I run it on my Windows Server 2008, 24 cores, 160G RAM. Even I set hash to 131072, the taskmgr said Sting only allocated 8G RAM, and CPU does not keep on 100%, unlike stockfish
Sting is based on the old version (2.1.1) of Stockfish.
Indeed, Sting can use "only" 8G (max) of Hash but I don't know: is much bigger (than 1024M) hash able to help something, or not?
It really helps when you need deep analysis. Hash stores evaluated positions. When the hash is nearly full after a long-time analysis, old positions will be replaced by new ones. However, the next time the replaced position is visited, the engine will have to re-evaluate it. That is to say, the larger the hash is, the less repetitive work the engine will do.
It is only well known part (good side) of the hash size theory. :)
Maybe, I can't be friendly, but let me be useful.

Post Reply