Why McBrain?

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

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Dann Corbit
Posts: 9067
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Why McBrain?

Post by Dann Corbit » Sat Sep 02, 2017 5:57 am

I want the right answer.
I use chess engines for analysis much more than for game play.
I don't want a winning move.
I want the best move.
And I want the whole plan.

Call me greedy. I have no problem with that.

Someone might offer:
It won't make the engine get more Elo to find the optimal path.
My answer:
Once a winning path is found, there is nothing lost by finding the best plan.

For people who just want an engine that buzzes away online to build up a big Elo score, there is no need. But if someone wants an engine to analyze games and positions, then there is a need.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

peter
Posts: 1629
Joined: Sat Feb 16, 2008 6:38 am

Re: Why McBrain?

Post by peter » Sat Sep 02, 2017 6:09 am

Thanks again, Michael!

I like the new White's POV Respect option and the nex maximum of 300 too.
:)

But it's human nature to want more the more you get, so what do you think about Persistent Hash?

Is a "Never Clear Hash" option accordingly helpful or even necessary?

Or is the way of SF- PA_GTB learning file the more interesting way to go?

And to be fully unhumble: could you imagine a Shared Hash- feature between two SF- Settings once upon a time
:?:
Peter Martan

mcostalba
Posts: 2679
Joined: Sat Jun 14, 2008 7:17 pm

Re: Why McBrain?

Post by mcostalba » Sat Sep 02, 2017 8:29 am

Stockfish Natural TB

Code: Select all

$ ./stockfish
Stockfish 020917 64 POPCNT by T. Romstad, M. Costalba, J. Kiiski, G. Linscott
info string Found 145 tablebases
position fen 1r4rk/4Nq2/7K/8/8/6R1/b5Q1/6R1 w - - 0 1
d

 +---+---+---+---+---+---+---+---+
 |   | r |   |   |   |   | r | k |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   | N | q |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   | K |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   | R |   |
 +---+---+---+---+---+---+---+---+
 | b |   |   |   |   |   | Q |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   | R |   |
 +---+---+---+---+---+---+---+---+

Fen: 1r4rk/4Nq2/7K/8/8/6R1/b5Q1/6R1 w - - 0 1
Key: B472FE7C776AC07F
Checkers:
go depth 12
info depth 1 seldepth 2 multipv 1 score cp 978 nodes 113 nps 113000 tbhits 0 time 1 pv g3g8 b8g8 g2b2 g8g7 g1g7
info depth 2 seldepth 4 multipv 1 score cp 715 nodes 231 nps 231000 tbhits 0 time 1 pv g3g8 f7g8 e7g8
info depth 3 seldepth 5 multipv 1 score cp 978 nodes 371 nps 371000 tbhits 0 time 1 pv g3g8 b8g8 g2b2 g8g7 g1g7
info depth 4 seldepth 8 multipv 1 score cp 1579 nodes 701 nps 701000 tbhits 2 time 1 pv g3g8 f7g8 e7g8 b8b6 h6h5 b6g6 g2g6
info depth 5 seldepth 7 multipv 1 score cp 1579 nodes 808 nps 808000 tbhits 4 time 1 pv g3g8 f7g8 e7g8 b8b6 h6h5 b6g6 g2g6
info depth 6 seldepth 8 multipv 1 score cp 13277 nodes 1024 nps 1024000 tbhits 22 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8
info depth 7 seldepth 6 multipv 1 score cp 13277 nodes 1164 nps 1164000 tbhits 38 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8
info depth 8 seldepth 8 multipv 1 score cp 13277 nodes 1381 nps 1381000 tbhits 64 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8
info depth 9 seldepth 8 multipv 1 score cp 13277 nodes 1690 nps 1690000 tbhits 88 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8
info depth 10 seldepth 10 multipv 1 score mate 5 nodes 2460 nps 2460000 tbhits 173 time 1 pv g3g8 f7g8 e7g6 g8g6 g2g6 b8b6 g6b6 a2b3 b6d4
info depth 11 seldepth 10 multipv 1 score mate 4 nodes 3109 nps 3109000 tbhits 205 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8 a2g8 e7g6
info depth 12 seldepth 8 multipv 1 score mate 4 nodes 3495 nps 3495000 tbhits 209 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8 a2g8 e7g6
bestmove g3g8 ponder f7g8
ucinewgame
position fen 1r5k/4NP1b/7K/8/6R1/8/8/8 w - - 0 1
d

 +---+---+---+---+---+---+---+---+
 |   | r |   |   |   |   |   | k |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   | N | P |   | b |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   | K |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   | R |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+

Fen: 1r5k/4NP1b/7K/8/6R1/8/8/8 w - - 0 1
Key: B3D6B247200E7BAC
Checkers:
go depth 12
info depth 1 seldepth 2 multipv 1 score cp 237 nodes 49 nps 49000 tbhits 0 time 1 pv e7g6 h7g6 h6g6
info depth 2 seldepth 3 multipv 1 score cp 938 nodes 114 nps 114000 tbhits 0 time 1 pv e7c6 h7c2 c6b8
info depth 3 seldepth 3 multipv 1 score cp 920 nodes 163 nps 163000 tbhits 0 time 1 pv e7c6 h7d3 c6b8
info depth 4 seldepth 4 multipv 1 score cp 5098 nodes 369 nps 369000 tbhits 0 time 1 pv e7c6 h7g6 h6g6 b8g8 f7g8q h8g8
info depth 5 seldepth 6 multipv 1 score cp 13278 nodes 463 nps 463000 tbhits 6 time 1 pv e7g6 h7g6 g4g6
info depth 6 seldepth 4 multipv 1 score cp 13278 nodes 504 nps 504000 tbhits 7 time 1 pv e7g6 h7g6 g4g6
info depth 7 seldepth 4 multipv 1 score cp 13278 nodes 545 nps 545000 tbhits 8 time 1 pv e7g6 h7g6 g4g6
info depth 8 seldepth 6 multipv 1 score mate 3 nodes 736 nps 736000 tbhits 32 time 1 pv g4g8 b8g8 f7g8r h7g8 e7g6
info depth 9 seldepth 6 multipv 1 score mate 3 nodes 866 nps 866000 tbhits 39 time 1 pv g4g8 b8g8 f7g8r h7g8 e7g6
info depth 10 seldepth 6 multipv 1 score mate 3 nodes 1026 nps 21375 tbhits 49 time 48 pv g4g8 b8g8 f7g8r h7g8 e7g6
info depth 11 seldepth 6 multipv 1 score mate 3 nodes 1187 nps 24729 tbhits 56 time 48 pv g4g8 b8g8 f7g8r h7g8 e7g6
info depth 12 seldepth 6 multipv 1 score mate 3 nodes 1442 nps 30041 tbhits 65 time 48 pv g4g8 b8g8 f7g8r h7g8 e7g6
bestmove g4g8 ponder b8g8
BTW according to my tests with 5-men it is also way stronger than current version:

Code: Select all

Stockfish NTB vs Stockfish master:  2143 - 1736 - 6121 [0.520]
Elo difference: 14.15  +/- 4.23
10000 of 10000 games finished.

P.S: Michael, I fully support your fork and I hope you will continue with McBrain and many will download and try it.

Ras
Posts: 1021
Joined: Tue Aug 30, 2016 6:19 pm
Contact:

Re: Why McBrain?

Post by Ras » Sat Sep 02, 2017 10:33 am

CT800 V1.12-pre (current dev version):

[d]1r4rk/4Nq2/7K/8/8/6R1/b5Q1/6R1 w - -

Code: Select all

  3/05	 0:00 	+5.32 	1.Rg3xg8+ Rb8xg8 2.Qg2-b2+ Rg8-g7 3.Rg1xg7 (10.941) 51  TB:4.715
  4/07	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (11.765) 54  TB:5.050
  5/07	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (12.435) 57  TB:5.521
  6/07	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (17.623) 80  TB:8.178
  7/07	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (31.116) 135  TB:14.017
  8/08	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (37.951) 160  TB:15.991
  9/09	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (67.442) 262  TB:26.068
 10/10	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (100.423) 358  TB:36.767
 11/11	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (193.840) 550  TB:78.876
 12/12	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (400.842) 782  TB:171.672
 13/13	 0:00 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (786.229) 995  TB:341.325
 14/14	 0:01 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (1.356.631) 1150  TB:569.040
 15/15	 0:01 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (2.319.676) 1285  TB:965.097
 16/16	 0:02 	 +M4 	1.Rg3xg8+ Rb8xg8 2.Qg2xg8+ Qf7xg8 3.Rg1xg8+ Ba2xg8 4.Ne7-g6+ (4.127.865) 1394  TB:1.710.555
[d]1r5k/4NP1b/7K/8/6R1/8/8/8 w - - 0 1

Code: Select all

  3/05	 0:00 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (9.070) 42  TB:3.715
  4/05	 0:00 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (10.561) 49  TB:4.601
  5/05	 0:00 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (11.961) 55  TB:4.993
  6/06	 0:00 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (19.990) 90  TB:9.275
  7/07	 0:00 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (43.228) 183  TB:20.297
  8/08	 0:00 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (93.371) 349  TB:45.620
  9/09	 0:00 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (185.067) 567  TB:94.828
 10/10	 0:00 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (333.421) 780  TB:175.294
 11/11	 0:00 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (706.529) 1057  TB:400.933
 12/12	 0:01 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (1.447.815) 1252  TB:795.849
 13/13	 0:02 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (3.025.982) 1405  TB:1.697.708
 14/14	 0:04 	 +M3 	1.Rg4-g8+ Rb8xg8 2.f7xg8Q+ Bh7xg8 3.Ne7-g6+ (6.315.022) 1481  TB:3.302.782

Rodolfo Leoni
Posts: 544
Joined: Tue Jun 06, 2017 2:49 pm
Location: Italy

Re: Why McBrain?

Post by Rodolfo Leoni » Sat Sep 02, 2017 12:42 pm

MikeB wrote:...............................
Just a general comment not directed at anybody -Why are People are so sensitive around here? You post something the you thought would be not offensive to anybody - yet somebody gets offended by it. Or you post something that is not 100% correct but for simplicity you used plainer language and people have the need to point it out but it really does not change the point one was making. Way too much drama for me for something that is supposed to be a hobby.
Maybe, you're considered a leader and it's expected everything you write is 100% true according to people's expectatives. I't probably true your work is interesting the most and everybody want to find absolute perfection (which could never exist). So you'll find those who want a perfect analysis engine because they need it, those who want perfect correspondence chess engine (like me) with Daniel's code (which is possible when you decide IMO), those who like the fact to have a top engine which "learns" (Ouch! Can a programmer give a soul to its engine?).

There's only an answer. Continue with your excellent work the way you like it. It's YOUR branch.
:)
F.S.I. Chess Teacher

carldaman
Posts: 1588
Joined: Sat Jun 02, 2012 12:13 am

Re: Why McBrain?

Post by carldaman » Sat Sep 02, 2017 12:53 pm

Rodolfo Leoni wrote:
MikeB wrote:...............................
Just a general comment not directed at anybody -Why are People are so sensitive around here? You post something the you thought would be not offensive to anybody - yet somebody gets offended by it. Or you post something that is not 100% correct but for simplicity you used plainer language and people have the need to point it out but it really does not change the point one was making. Way too much drama for me for something that is supposed to be a hobby.
Maybe, you're considered a leader and it's expected everything you write is 100% true according to people's expectatives. I't probably true your work is interesting the most and everybody want to find absolute perfection (which could never exist). So you'll find those who want a perfect analysis engine because they need it, those who want perfect correspondence chess engine (like me) with Daniel's code (which is possible when you decide IMO), those who like the fact to have a top engine which "learns" (Ouch! Can a programmer give a soul to its engine?).

There's only an answer. Continue with your excellent work the way you like it. It's YOUR branch.
:)
A big thanks to Mike B for his A+ efforts. :)

I do second Rodolfo's request/suggestion to add Daniel's code for analysis in the future. I have mostly used DJ's Stockfish versions since he kindly released them to us. The back-and-forth analysis is of so much better quality with the new method.

A major 'thank you' is in order to Daniel Jose as well, of course, and to Rodolfo for pointing out the original problem that DJ fixed. :)

Cheers,
CL

Rodolfo Leoni
Posts: 544
Joined: Tue Jun 06, 2017 2:49 pm
Location: Italy

Re: Why McBrain?

Post by Rodolfo Leoni » Sat Sep 02, 2017 2:07 pm

carldaman wrote: A big thanks to Mike B for his A+ efforts. :)

I do second Rodolfo's request/suggestion to add Daniel's code for analysis in the future. I have mostly used DJ's Stockfish versions since he kindly released them to us. The back-and-forth analysis is of so much better quality with the new method.

A major 'thank you' is in order to Daniel Jose as well, of course, and to Rodolfo for pointing out the original problem that DJ fixed. :)

Cheers,
CL
Off topic here, but I'm about to test "CorCFish extended" with correspondence games.

https://github.com/amchess/CorCfishExte ... xecutables

With Daniel's code. :)
F.S.I. Chess Teacher

User avatar
Eelco de Groot
Posts: 3975
Joined: Sun Mar 12, 2006 1:40 am
Location: Groningen

Re: Why McBrain?

Post by Eelco de Groot » Sat Sep 02, 2017 2:28 pm

mcostalba wrote:Stockfish Natural TB

Code: Select all

$ ./stockfish
Stockfish 020917 64 POPCNT by T. Romstad, M. Costalba, J. Kiiski, G. Linscott
info string Found 145 tablebases
position fen 1r4rk/4Nq2/7K/8/8/6R1/b5Q1/6R1 w - - 0 1
d

 +---+---+---+---+---+---+---+---+
 |   | r |   |   |   |   | r | k |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   | N | q |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   | K |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   | R |   |
 +---+---+---+---+---+---+---+---+
 | b |   |   |   |   |   | Q |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   | R |   |
 +---+---+---+---+---+---+---+---+

Fen: 1r4rk/4Nq2/7K/8/8/6R1/b5Q1/6R1 w - - 0 1
Key: B472FE7C776AC07F
Checkers:
go depth 12
info depth 1 seldepth 2 multipv 1 score cp 978 nodes 113 nps 113000 tbhits 0 time 1 pv g3g8 b8g8 g2b2 g8g7 g1g7
info depth 2 seldepth 4 multipv 1 score cp 715 nodes 231 nps 231000 tbhits 0 time 1 pv g3g8 f7g8 e7g8
info depth 3 seldepth 5 multipv 1 score cp 978 nodes 371 nps 371000 tbhits 0 time 1 pv g3g8 b8g8 g2b2 g8g7 g1g7
info depth 4 seldepth 8 multipv 1 score cp 1579 nodes 701 nps 701000 tbhits 2 time 1 pv g3g8 f7g8 e7g8 b8b6 h6h5 b6g6 g2g6
info depth 5 seldepth 7 multipv 1 score cp 1579 nodes 808 nps 808000 tbhits 4 time 1 pv g3g8 f7g8 e7g8 b8b6 h6h5 b6g6 g2g6
info depth 6 seldepth 8 multipv 1 score cp 13277 nodes 1024 nps 1024000 tbhits 22 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8
info depth 7 seldepth 6 multipv 1 score cp 13277 nodes 1164 nps 1164000 tbhits 38 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8
info depth 8 seldepth 8 multipv 1 score cp 13277 nodes 1381 nps 1381000 tbhits 64 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8
info depth 9 seldepth 8 multipv 1 score cp 13277 nodes 1690 nps 1690000 tbhits 88 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8
info depth 10 seldepth 10 multipv 1 score mate 5 nodes 2460 nps 2460000 tbhits 173 time 1 pv g3g8 f7g8 e7g6 g8g6 g2g6 b8b6 g6b6 a2b3 b6d4
info depth 11 seldepth 10 multipv 1 score mate 4 nodes 3109 nps 3109000 tbhits 205 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8 a2g8 e7g6
info depth 12 seldepth 8 multipv 1 score mate 4 nodes 3495 nps 3495000 tbhits 209 time 1 pv g3g8 f7g8 g2g8 b8g8 g1g8 a2g8 e7g6
bestmove g3g8 ponder f7g8
ucinewgame
position fen 1r5k/4NP1b/7K/8/6R1/8/8/8 w - - 0 1
d

 +---+---+---+---+---+---+---+---+
 |   | r |   |   |   |   |   | k |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   | N | P |   | b |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   | K |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   | R |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+
 |   |   |   |   |   |   |   |   |
 +---+---+---+---+---+---+---+---+

Fen: 1r5k/4NP1b/7K/8/6R1/8/8/8 w - - 0 1
Key: B3D6B247200E7BAC
Checkers:
go depth 12
info depth 1 seldepth 2 multipv 1 score cp 237 nodes 49 nps 49000 tbhits 0 time 1 pv e7g6 h7g6 h6g6
info depth 2 seldepth 3 multipv 1 score cp 938 nodes 114 nps 114000 tbhits 0 time 1 pv e7c6 h7c2 c6b8
info depth 3 seldepth 3 multipv 1 score cp 920 nodes 163 nps 163000 tbhits 0 time 1 pv e7c6 h7d3 c6b8
info depth 4 seldepth 4 multipv 1 score cp 5098 nodes 369 nps 369000 tbhits 0 time 1 pv e7c6 h7g6 h6g6 b8g8 f7g8q h8g8
info depth 5 seldepth 6 multipv 1 score cp 13278 nodes 463 nps 463000 tbhits 6 time 1 pv e7g6 h7g6 g4g6
info depth 6 seldepth 4 multipv 1 score cp 13278 nodes 504 nps 504000 tbhits 7 time 1 pv e7g6 h7g6 g4g6
info depth 7 seldepth 4 multipv 1 score cp 13278 nodes 545 nps 545000 tbhits 8 time 1 pv e7g6 h7g6 g4g6
info depth 8 seldepth 6 multipv 1 score mate 3 nodes 736 nps 736000 tbhits 32 time 1 pv g4g8 b8g8 f7g8r h7g8 e7g6
info depth 9 seldepth 6 multipv 1 score mate 3 nodes 866 nps 866000 tbhits 39 time 1 pv g4g8 b8g8 f7g8r h7g8 e7g6
info depth 10 seldepth 6 multipv 1 score mate 3 nodes 1026 nps 21375 tbhits 49 time 48 pv g4g8 b8g8 f7g8r h7g8 e7g6
info depth 11 seldepth 6 multipv 1 score mate 3 nodes 1187 nps 24729 tbhits 56 time 48 pv g4g8 b8g8 f7g8r h7g8 e7g6
info depth 12 seldepth 6 multipv 1 score mate 3 nodes 1442 nps 30041 tbhits 65 time 48 pv g4g8 b8g8 f7g8r h7g8 e7g6
bestmove g4g8 ponder b8g8
BTW according to my tests with 5-men it is also way stronger than current version:

Code: Select all

Stockfish NTB vs Stockfish master:  2143 - 1736 - 6121 [0.520]
Elo difference: 14.15  +/- 4.23
10000 of 10000 games finished.

P.S: Michael, I fully support your fork and I hope you will continue with McBrain and many will download and try it.
There is a -.bmi compile available of Marco's Natural TB, I got the sources that I think belong to Marco's testresults and compiled with TDM-GCC, now on Rybka forum, but it has some eval changes from me included, which are not yet tested with nearly the same accuracy. Only 50 games Still, if Marco's resuts are even remotely reliable, this very probably would be about as strong as asmFish and if you add Jeroen Noomens book... :D Compile is called Natural.exe

Only 1130 kb! Bench:

===========================
Total time (ms) : 2527
Nodes searched : 4974113
Nodes/second : 1968386
TB hits : 0
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

syzygy
Posts: 4343
Joined: Tue Feb 28, 2012 10:56 pm

Re: Why McBrain?

Post by syzygy » Sat Sep 02, 2017 2:50 pm

Eelco de Groot wrote:if Marco's resuts are even remotely reliable
Don't count on it. I don't know what went wrong, but there is no way that his patch can improve playing strength (and certainly not by that amount).

In the test I am running, the early_mate patch (which is now in McBrain) is very slightly ahead of natural2. The early_mate patch certainly does not add Elo. But it seems to do the job of finding a mate "behind" a TB win quite well (not worse than natural2, is my impression when browsing through the games).

As a "bonus", early_mate keeps all the guarantees that the current TB implementation provides, such as the ability to reliably convert any TB win found.

Once Marco has merged his so-called "natural TB" patch, SF will stop doing well, in particular, on the positions discussed here:
http://talkchess.com/forum/viewtopic.php?t=59237

giovanni
Posts: 104
Joined: Wed Jul 08, 2015 10:30 am

Re: Why McBrain?

Post by giovanni » Sat Sep 02, 2017 3:13 pm

Rodolfo Leoni wrote:
carldaman wrote: A big thanks to Mike B for his A+ efforts. :)

I do second Rodolfo's request/suggestion to add Daniel's code for analysis in the future. I have mostly used DJ's Stockfish versions since he kindly released them to us. The back-and-forth analysis is of so much better quality with the new method.

A major 'thank you' is in order to Daniel Jose as well, of course, and to Rodolfo for pointing out the original problem that DJ fixed. :)

Cheers,
CL
Off topic here, but I'm about to test "CorCFish extended" with correspondence games.

https://github.com/amchess/CorCfishExte ... xecutables
y
With Daniel's code. :)
Thanks Rodolfo. Sounds interesting. Could you detail a little more what you are doing with this project?

Post Reply