Honey 13.1 is around the corner ...

[MikeB]

Doing Final Testing at roughly 2 min + 1 ( just to make sure no thing is really out of wack, closed the watcher program by accident , so there will be no live updates, but you can grab the games if you wish from dropbox -> https://www.dropbox.com/s/022ilbv3btk3r ... 0.pgn?dl=0 - 2700 games are scheduled ( will wrap up around around 9 AM EDST) The Honey engines will all come with a new, never released net, nn-811243cea2cf.nnue , the same net that is doing so well solving those hard positions when paired with Black Diamond.

The stockfish below is Stockfish 13, not current -dev. SF dominates the Honey engines by at least 7 to 10 elo or more at faster time controls and it is probably true here too,
I just run this last test to make sure nothing is really broken bad with the Honey engines, It's good to see Black Diamond within 45 Elo or so of SF.

Code: Select all

Rank Name                             Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR
----------------------------------------------------------------------------------------------------------------
   1 Stockfish-62ef826d1a6d            3120   0.0   13   13   136   74.0  54.4   16    4  116  11.8  85.3  3096
   2 Oki-Maguro-13.1-811243cea2cf      3113   6.7   14   14   135   71.5  53.0   13    5  117   9.6  86.7  3097
   3 Honey-13.1-811243cea2cf           3110   3.1   14   14   136   71.0  52.2   11    5  120   8.1  88.2  3098
   4 Harmon-13.1-811243cea2cf          3094  16.0   14   14   135   66.0  48.9    3    6  126   2.2  93.3  3101
   5 Bluefish-13.1-811243cea2cf        3093   1.6   14   14   134   65.0  48.5    4    8  122   3.0  91.0  3102
   6 Black-Diamond-13.1-811243cea2cf   3070  23.1   12   12   134   57.5  42.9    1   20  113   0.7  84.3  3106
----------------------------------------------------------------------------------------------------------------

==

If all goes well, it might be Saturday. Harmon Chess App ( a modified Droidfish) will be updated at the same time.
A new feature being added to the Harmon Chess App is the ability to play games against the Honey engines using search by depth (up to depth 25) or search by nodes (up to 100,000) and you can combine those two options then as well, it will play which ever one hits first. You will want to turn the sleep the function if you not want the engine to move instantly Purechess suggested it and it was a great suggestion, pretty slick. Brings a new option to play your favorite engine. As an example search depth 1 would be an excellent setting for beginners.

Edit: Just as a reminder, Black Diamond has Fortress detection code by Joe Ellis and it will find ways to crack through a fortress on occasion,

Example:
[d]3r4/3r4/2p1k3/p1pbPp1p/Pp1p1PpP/1P1P2R1/2P4R/2K1Q3 w - - 0 1

single thread

Code: Select all

 32	+3.45!	407.7M	4:05.14	Rxg4!
 32	+3.42!	393.7M	3:56.92	Rxg4!
 31	+3.48!	368.4M	3:42.82	Rxg4!
 31	+3.40!	328.4M	3:19.47	Rxg4!
 31	+3.36?	327.9M	3:19.14	Rxg4 hxg4?
 31	+3.41?	327.7M	3:19.03	Rxg4 hxg4?
 31	+3.46!	298.8M	3:02.04	Rxg4!
 30	+3.37 	285.4M	2:54.09	Rxg4 hxg4 h5 Rh8 Qh4 Rf7 Kb2 Kd7 h6 Kc8 Qg5 Kb7 Rh5 Ka6 Qg6 Rfh7 e6 Kb6 Rh4 Bf3 Rh2 Re7 h7 Bd5 Qf6 Rexh7 Rxh7 Rxh7 e7 Rf7 Qd6 Rxe7 Qxe7 g3 Qd8+ Kb7 Qd7+ Ka8 Qc8+ Ka7 Qxf5 g2 Qh7+ Ka6 Qg7 c4 dxc4
 30	+3.36!	283.8M	2:53.14	Rxg4!
 30	+3.27?	283.8M	2:53.08	Rxg4 hxg4?
 30	+3.39!	267.3M	2:43.39	Rxg4!
 30	+3.35!	248.3M	2:31.97	Rxg4!
 30	+3.33!	239.4M	2:26.48	Rxg4!
 29	+3.23 	226.2M	2:18.54	Rxg4 hxg4 h5 Rg8 Qh4 Rf7 h6 Kd7 h7 Rh8 Qg5 Kc7 Qg8 g3 Qxg3 Rhxh7 Rxh7 Rxh7 Qg6 Rf7 Qd6+ Kb6 e6 Bxe6 Qxe6 Rc7 Qe5 c4 bxc4 Kb7 Qxa5 c5 Qb5+ Kc8 a5 Kd8
 29	+3.22!	198.4M	2:02.23	Rxg4!
 29	+3.13?	197.4M	2:01.59	Rxg4 hxg4?
 29	+3.25!	170.5M	1:45.76	Rxg4!
 29	+3.21!	147.6M	1:32.12	Rxg4!
 29	+3.18!	121.6M	1:16.16	Rxg4!
 28	+2.93 	117.1M	1:13.43	Rxg4 hxg4 h5 Rh7 Qh4 Kd7 e6+ Bxe6 Re2 Bd5 Re5 Rf7 h6 Rh8 Qg5 g3 Qxg3 Rxh6 Qg8 Rhh7 Qa8 Re7 Rxe7+ Rxe7 Qxa5 Bxb3 cxb3 c4 dxc4 Re3 Qxf5+ Kc7 Qa5+ Kd7 Qc5 Rxb3 Qxd4+ Kc7 Kc2 Rf3 a5 b3+ Kb2
 28	+2.97!	115.6M	1:12.62	Rxg4!
 28	+2.87!	93.7M  	0:59.24	Rxg4!
 28	+2.79!	85.4M  	0:54.07	Rxg4!
 28	+2.73!	83.0M  	0:52.58	Rxg4!
 28	+2.69?	82.8M  	0:52.45	Rxg4 hxg4?
 28	+2.74!	81.7M  	0:51.75	Rxg4!
 27	+2.72 	77.9M  	0:49.44	Rxg4 hxg4 h5 Rh8 Qh4 Rf7 Kb2 Kd7 h6 Kc8 Qg5 Kc7 Rh5 Kb6 Qg6 Rfh7 e6 g3 Qxg3 Bxe6 Qg5 Bf7 Rh3 Bd5 Qg6 Be4 dxe4 fxe4 Qxe4 Rxh6
 26	+2.74 	75.4M  	0:47.86	Rxg4 hxg4 h5 Rh8 Qh4 Rf7 Kb2 Kd7 h6 Kc8 Qg5 Kc7 Rh5 Kb6 Qg6 Rfh7 e6 g3 Qxg3 Bxe6 Qg5 Bf7 Rh2 Bd5 Qg6 Bf3 Qf6 Be4 dxe4 fxe4
 26	+2.75!	71.4M  	0:45.40	Rxg4!
 26	+2.65?	71.3M  	0:45.38	Rxg4 hxg4?
 26	+2.71?	67.3M  	0:42.85	Rxg4 hxg4?
 26	+2.79!	64.4M  	0:41.02	Rxg4!
 26	+2.77!	63.6M  	0:40.53	Rxg4!
 25	+2.67 	60.4M  	0:38.67	Rxg4 hxg4 h5 Rh7 Qh4 Kd7 e6+ Bxe6 Re2 Bd5 Re5 Rf7 h6 Rh8 Qh5 Rf6 Qg5 Rhxh6 Qg7+ Bf7 Rxc5 Rhg6 Qh8 g3 Rxa5 Bg8 Ra7+ Kd8 Ra8+ Ke7 Rxg8 Rxg8 Qxg8 c5 Qxg3
 25	+2.57?	60.4M  	0:38.64	Rxg4 hxg4?
 25	+2.75!	55.5M  	0:35.56	Rxg4!
 25	+2.70!	51.4M  	0:33.00	Rxg4!
 25	+2.65!	48.9M  	0:31.35	Rxg4!
 25	+2.63!	47.1M  	0:30.24	Rxg4!
 24	+2.51 	45.1M  	0:28.94	Rxg4 hxg4 h5 Rh7 Qh4 Kd7 e6+ Bxe6 Re2 Bd5 Re5 Rf8 Qg5 Kc8 h6 Rhf7 Kb1 Kb7 Re7+ Kb6 Rxf7 Rxf7 Qd8+ Ka6 Qg8 Rb7 Qg6 Kb6 h7 Rb8 Qxf5 g3
 24	+2.53!	43.9M  	0:28.27	Rxg4!
 24	+2.39?	41.6M  	0:26.81	Rxg4 hxg4?
 24	+2.47!	41.0M  	0:26.49	Rxg4!
 24	+2.45!	40.6M  	0:26.22	Rxg4!
 23	+2.40!	39.0M  	0:25.25	Rxg4!
 23	+2.36!	38.6M  	0:24.94	Rxg4!
 23	+2.33!	38.0M  	0:24.57	Rxg4!
 22	+2.25!	36.9M  	0:23.86	Rxg4!
 22	+2.21?	36.9M  	0:23.86	Rxg4 hxg4?
 22	+2.27!	35.5M  	0:22.96	Rxg4!
 21	+2.52!	34.3M  	0:22.20	Rxg4!
 21	+2.24!	33.8M  	0:21.88	Rxg4!
 21	+2.02!	33.6M  	0:21.74	Rxg4!
 21	+1.87?	33.5M  	0:21.66	Rxg4 hxg4?
 21	+2.06!	33.3M  	0:21.55	Rxg4!
 21	+1.98!	29.4M  	0:18.99	Rf2!
 21	+1.95?	26.8M  	0:17.41	Rg1 Rg8?
 20	+2.17!	21.3M  	0:13.89	Rg1!
 20	+2.13?	20.3M  	0:13.24	Rg1 Rg8?
 20	+2.19?	17.7M  	0:11.56	Rg1 Rg8?
 20	+2.23?	15.3M  	0:10.10	Rg1 Rg8?
 20	+2.25?	14.1M  	0:09.32	Rg1 Rg8?
 19	+2.28 	13.3M  	0:08.76	Rg1 Rg8 Kd2 Rf8 Ke2 Rfd8 Qg3 Rb8 Kf2 Rf8 Re1 Bf3 Kg1 Bd5 Rhe2 Bf3 Rf2 Bd5 Rff1 Rb7 Rc1 Rd7 Rfe1 Rb7 c4 bxc3 Qf2 Rxb3 Rc2 Rb4
 19	+2.36!	12.8M  	0:08.49	Rg1!
 19	+2.20!	10.4M  	0:06.86	Rf2!
 19	+2.08!	10.4M  	0:06.83	Rf2!
 19	+2.01?	9.90M  	0:06.53	Rxg4 hxg4?
 19	+2.11!	9.87M  	0:06.51	Rxg4!
 18	+2.06 	8.79M  	0:05.78	Re2 Rf8 Qd1 Rfd8 Qg1 Re8 Re1 Red8 Rge3 Rf8 Qg3 Rfd8 Re4 Rf8 Kb2 Rb7 Qg1 Rc8 Ra1 Rd7 Rf1 fxe4 f5+ Kxe5 Qg3+ Kf6 Qf4 exd3 cxd3 Rf7
 18	+2.21!	8.00M  	0:05.24	Re2!
 18	+2.05!	7.56M  	0:04.94	Re2!
 18	+2.08?	6.92M  	0:04.51	Kd1 Rf8?
 18	+2.12?	6.10M  	0:04.00	Kd1 Rf8?
 18	+2.14?	5.55M  	0:03.69	Kd1 Rf8?
 17	+2.17 	4.35M  	0:02.89	Kd1 Rf8 Re2 Rfd8 Qg1 Rf8 Ke1 Rfd8 Kf2 Rf8 Re1 Rfd8 Rb1 Rf8 Qd1 Rfd8 Qc1 Rb8 Rg1 Rbd8 c3 dxc3 Qe3 c4 bxc4 Bf3 Qc5 Rxd3
 16	+1.84 	3.44M  	0:02.29	Rxg4 hxg4 h5 Rg8 h6 Rh7 Qh4 Kd7 e6+ Bxe6 Qf6 Rf7 Qe5 Rh7 Qxc5 g3 Re2 Rxh6 Qxd4+ Kc8 Qd6 Kb7 Rxe6 Rxe6 Qxe6 Rg7 Qh6
 15	+1.88 	2.50M  	0:01.70	Rxg4 hxg4 h5 Rg8 Qh4 Rf7 h6 Kd7 Kd2 Be6 Qg3 Rh7 Rh5 Kd8 Kc1
 14	+2.21 	746967	0:00.51	Kd1 Rf8 Ke2 Rfd8 Kf1 Rb8 Qc1 Rdd8 Re2 Rd7 Kf2 Rdd8 Re1 Rb7 Rgg1 Rg8 Kg3 Rbb8 Kh2 Rgd8 Qb2 Rd7 Rb1
 13	+2.16 	538300	0:00.36	Kd1 Rf8 Ke2 Rg8 Kf1 Rc8 Re2 Rb8 Qc1 Rdd8 c3 bxc3
 12	+2.51 	325346	0:00.22	Rxg4 hxg4 h5 Rg8 Qh4 Rf7 h6 Kd7 h7 Rh8 Qg5 g3 Rh6 g2 e6+ Bxe6
 11	+2.64 	254019	0:00.16	Rxg4 hxg4 h5 Rg8 Qh4 Rf7 h6 Kd7 h7 Rh8 Qg5 g3 e6+ Bxe6
 10	+2.99 	119082	0:00.07	Kb1 Rf8 Kb2 Rfd8 Rxg4 hxg4 h5 Rc8 Qh4 Rf7 h6 Rh8 Qg5 Rhh7 Qg6+ Ke7
  9	+2.59 	91341  	0:00.06	Kb2 Rf8 Rg1 Bf3 Re2 Rfd8 Rd2 Bd5 Ka1
  8	+2.94 	58308  	0:00.04	Kb2 Rf8 Rg1 Rfd8 Re2 Rf8 Ka2 Bf3 Qf2 Bxe2 Qxe2 Rg8
  7	+2.93 	21250  	0:00.02	Kb2 Rf8 Re2 Rfd8 Rg1 Rf8 Rf1 Rfd8 Rd2 Rf8 Ka1 Rfd8 Qc1 Rf8 Qb2 Rfd8 Rdd1 Bf3 Rb1 Ra7 Rbc1
  6	+2.68 	9865    	0:00.01	Kb2 Rf8 Re2 Rfd8 Rg1 Rf8 Rf1 Rfd8 Qf2 Bf3 Rfe1 Bxe2 Qxe2
  5	+4.12 	3314    	0:00.00	Kb2 Rf8 Re2 Rfd8 Rg1 Rf8 Rf1 Rfd8 Qf2 Bf3 Ree1
  4	+4.10 	654      	0:00.00	Kb2 Rf8 Re2 Rfd8
  3	+4.12 	290      	0:00.00	Kb2 Rf8 Rg1
  2	+4.09 	163      	0:00.00	Kb2 Rf8
  1	NNUE evaluation using C:\cluster.mfb\Popcnt-LP\eval\nn-811243cea2cf.nnue enabled
  1	+4.04 	63        	0:00.00	Kb2
  0	#

[odyn1982]

Hi,

Can you add NNUE switch: Pure, Hybrid and Classical? I prefer to play against some neural nets with Pure option on.

Also "Variety" is very useful. Could you add it back?

[MikeB]

Can you add NNUE switch: Pure, Hybrid and Classical? I prefer to play against some neural nets with Pure option on.

Let me look , Classical is no longer as interesting since "some" of former classical eval it is now being handled by the net and some former classical code removed. So I'm thinking a pure mode check box will be good and is fairly simple to add. Pure mode is entertaining for humans to play against.

Also "Variety" is very useful. Could you add it back?

If you use ANY of the limit strength settings, they all have variety built in (by definition, variety is limiting strength).
But the new play by depth or play nodes searched has no variety, so I'll take a look. If not this release, I can add it back the next release.

[MikeB]

some test results, made some tweaks to Harmon
this was at 2 min + 1 sec inc.
Stockfish dominates here and is even more dominating at shorter tc. One loss in 130 games is amazing. These engines are not push overs for other engines.

Theses short runs can be a little flukey, but the fact that SF dominates here is not flukey. Places 2, 3, and 4 are easily interchangeable.

Code: Select all

Rank Name                             Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR
---------------------------------------------------------------------------------------------------------------

   1 Stockfish-62ef826d1a6d            3520   0.0   13   13   130   71.0  54.6   13    1  116  10.0  89.2  3496
   2 Harmon-13.1-811243cea2cf          3509  11.5   14   14   130   67.5  51.9   10    5  115   7.7  88.5  3498
   3 Bluefish-13.1-811243cea2cf        3505   3.5   14   14   130   66.5  51.2    8    5  117   6.2  90.0  3499
   4 Oki-Maguro-13.1-811243cea2cf      3501   3.8   14   14   130   65.5  50.4    6    5  119   4.6  91.5  3500
   5 Honey-13.1-811243cea2cf           3492   9.2   14   14   130   63.0  48.5    4    8  118   3.1  90.8  3502
   6 Black-Diamond-13.1-811243cea2cf   3472  20.1   12   12   130   56.5  43.5    1   18  111   0.8  85.4  3506
---------------------------------------------------------------------------------------------------------------

[purechess]

Great news....

Can you provide more info about the network nn-811243cea2cf.nnue ?

[Rowen]

Can you add NNUE switch: Pure, Hybrid and Classical? I prefer to play against some neural nets with Pure option on.

Let me look , Classical is no longer as interesting since "some" of former classical eval it is now being handled by the net and some former classical code removed. So I'm thinking a pure mode check box will be good and is fairly simple to add. Pure mode is entertaining for humans to play against.

Also "Variety" is very useful. Could you add it back?

If you use ANY of the limit strength settings, they all have variety built in (by definition, variety is limiting strength).
But the new play by depth or play nodes searched has no variety, so I'll take a look. If not this release, I can add it back the next release.

Hello
Thanks for releasing these useful stockfish versions. Slightly embarrassing request as it reveals my chess level, but I was wondering if it was possible to release a version that could take the nodes down to 1 node per second (as is possible with the Rodent engines) the stockfish versions are in multiples of 1000 nodes. This is still(slightly) to strong for patzers like myself. I realise that there are other ways of weakening the engines , and weakfish etc and this may not be a trivial change or even possibly worthwhile, so not important, but just thought i would ask.

Once again thanks for your versions which are always helpful and useful for playing against,

[MikeB]

Can you add NNUE switch: Pure, Hybrid and Classical? I prefer to play against some neural nets with Pure option on.

Let me look , Classical is no longer as interesting since "some" of former classical eval it is now being handled by the net and some former classical code removed. So I'm thinking a pure mode check box will be good and is fairly simple to add. Pure mode is entertaining for humans to play against.

Also "Variety" is very useful. Could you add it back?

If you use ANY of the limit strength settings, they all have variety built in (by definition, variety is limiting strength).
But the new play by depth or play nodes searched has no variety, so I'll take a look. If not this release, I can add it back the next release.

Hello
Thanks for releasing these useful stockfish versions. Slightly embarrassing request as it reveals my chess level, but I was wondering if it was possible to release a version that could take the nodes down to 1 node per second (as is possible with the Rodent engines) the stockfish versions are in multiples of 1000 nodes. This is still(slightly) to strong for patzers like myself. I realise that there are other ways of weakening the engines , and weakfish etc and this may not be a trivial change or even possibly worthwhile, so not important, but just thought i would ask.

Once again thanks for your versions which are always helpful and useful for playing against,

In this new version, set "play_depth" to 1 and "sleep" to 30 seconds and it will average about 1 nps.
With "play_depth" set to 2 , you will have to set "sleep" to 2 minutes to average about 1 nps.

[MikeB]

running behind, adding PureNN mode
it's looking sweet in some positions...
[d]2b1rk2/5p2/p1P5/2p2P2/2p5/7B/P7/2KR4 w - - 0 1
single thread 256 hash
BD 13.1

Code: Select all

	mover viewpoint		fewer / Multi-PV setting = 1 / more
exclude: none best +tail                                          
dep	score	nodes	time	(not shown:  tbhits	knps	seldep)
 27	+5.10!	115.4M	0:53.45	f6!
 27	+4.82!	105.1M	0:49.36	f6!
 27	+4.43!	98.8M  	0:46.79	f6!
 27	+4.35!	94.9M  	0:45.17	f6!
 27	+4.29!	91.7M  	0:43.87	f6!
 27	+4.25!	88.3M  	0:42.45	f6!
 27	+4.22?	87.8M  	0:42.23	f6 a5?
 26	+4.25!	75.5M  	0:37.45	f6!
 26	+3.27!	62.7M  	0:31.94	f6!
 26	+2.49!	52.5M  	0:26.94	f6!
 26	+1.88!	44.4M  	0:22.88	f6!
 26	+1.41!	40.6M  	0:20.93	f6!
 26	+1.04!	37.5M  	0:19.40	f6!
 26	+0.76!	35.6M  	0:18.42	f6!
 26	+0.54!	34.8M  	0:18.01	f6!
 26	+0.38!	33.7M  	0:17.50	f6!
 26	+0.26!	32.7M  	0:17.00	f6!
 26	+0.18!	32.0M  	0:16.65	f6!
 26	+0.13!	30.5M  	0:15.88	f6!
 25	+0.10 	20.5M  	0:10.64	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Ra1 Rd8 Bxf5 Kc3 Rc1+ Kd2 Rc2+ Ke3 Rc4 Bf3 Rb4 c8=Q Bxc8 Rxc8 Kd7 Ra8 Kd6 Rxa6+ Ke5 Ra3 Rb2 Bd1 Kd6
 25	+0.09?	20.3M  	0:10.54	c7 f6?
 24	+0.10!	10.5M  	0:05.56	c7!
 23	+0.07 	10.1M  	0:05.30	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Ra1 Rd8 Bxf5 Kc3 Rc1+ Kd2 Rc2+ Ke3 Rc4 Bf3 Rb4 c8=Q Bxc8 Rxc8 f5 Kd3 Kd7 Rxc5
 23	+0.09!	6.39M  	0:03.40	c7!
 22	+0.07 	5.44M  	0:02.89	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Ra1 Rd8 Bxf5 c8=Q Bxc8 Rxc8 Kd6 Be4 Ra4+ Kd3 f5 Ba8 c4+ Kd4
 21	+0.08 	4.02M  	0:02.12	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Ra1 Rd8 Bxf5 Kb3 Rb1+ Ka2 Rb6 c8=Q Bxc8 Rxc8 Kd6 Ba4 f5 Bb3 f4 Rd8+ Kc6
 20	+0.06 	3.69M  	0:01.95	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Ra1 Kb3 Rb1+ Ka3 Rb6 Ba4 Rb7 Rxc5 Kd6 Rc6+ Ke7 Bb3 Kd7 Rxf6 Kxc7 Rf7+ Kd6 Rxb7 Bxb7
 19	+0.06 	3.30M  	0:01.76	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Ra1 Kb3 Rb1+ Ka3 Rb6 Ba4 Rb7 Rxc5 Rb6 Rc2 Bxf5 Rc3 Bc8
 18	+0.11 	1.96M  	0:01.04	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Ra1 Rd8 Bxf5 Kxc5 Rc1+ Kb6 Ra1 Be4 Bd7 Ra8 Kd6 Re8 f5
 17	+0.16 	1.52M  	0:00.81	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Rc2+ Kb3 Re2 Rd8 Bxf5 c8=R Bxc8 Rxc8 Kd6 Ba4 f5 Rc6+ Ke5 Rxc5+ Kd6
 16	+0.13 	1.08M  	0:00.58	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Rc2+ Kb3 Re2 Rd8 Bxf5 c8=R Bxc8 Rxc8 Kd6 Ba4 Re4 Ka3 Rh4 Bb3 Re4
 15	+0.09 	871215	0:00.47	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Kc3 Rxa2 Kxc4 Rc2+ Kb3 Rc1 Rd8 Bxf5 Bd5 Rd1 Kc3 Be6 Bxe6 Rxd8 c8=Q
 14	+0.10 	552982	0:00.30	c7 f6 Bg2 Ke7 Bc6 Rg8 Rd5 Rg1+ Kb2 Rg2+ Ka3 Rg3+ Ka4 Rg2 a3 Rg8 Rxc5 Kd6 Kb4 Rg7 Rxc4 Rxc7
 13	+0.09 	442658	0:00.24	c7 f6 Bg2 Ke7 Kb2 Rg8 Bd5 Rh8 Bxc4 Rh4 Bb3 Rd4 Rxd4
 12	+0.12 	305880	0:00.17	Rd5 f6 c7 Ke7 Bg2 Rg8 Bf3 Bd7 Be4 Rc8 Rxc5 Kd6 Rxc4 Rxc7
 11	+0.15 	180105	0:00.10	Rd5 f6 Kd2 Re4 Bf1 Re5 Rxe5 fxe5 Bxc4 Bxf5
 10	+0.26 	75045  	0:00.04	c7 f6 Rd5 Ke7 Bg2 Rg8 Be4 Rg4 Bc2 Rd4 Rxc5 Kd6
  9	+0.38 	34154  	0:00.02	c7 f6 Rd5 Ke7 Bg4 Rh8 Rxc5 Kd6 Rxc4 Re8
  8	+0.42 	19221  	0:00.01	c7 f6 Bf1 Ke7 a3 c3 Kc2 Rh8 Kxc3
  7	+0.45 	14282  	0:00.01	Rd5 Re3 Bg4 Rd3 Rxc5 Ke7 Rxc4
  6	+0.50 	6437    	0:00.01	c7 Ke7 Rd5 Rh8 f6+ Kxf6
  5	+0.30 	1960    	0:00.00	Kc2 Ke7 Re1+ Kf8 Rxe8+ Kxe8 Kc3 Kd8 Kxc4
  4	+0.12 	941      	0:00.00	Kc2 Ke7 c7 Rh8
  3	+0.42 	317      	0:00.00	c7 Ke7 Kb2
  2	+0.34 	129      	0:00.00	c7 Ke7
  1	NNUE evaluation using C:\cluster.mfb\Popcnt-LP\eval\nn-811243cea2cf.nnue enabled
  1	+0.40 	32        	0:00.00	c7
  0	#
{/code]

[MikeB]

Great news....

Can you provide more info about the network nn-811243cea2cf.nnue ?

Note: What follows is a 60 second elevator speech , it may not be precisely accurate, but it is in plain language that I hope someone that is not familiar with the process will be able to digest.

The weights of a neural network cannot be calculated analytically, they are discovered , similarly to how bitcoin mining hash keys are also discovered. With both processes, you have a dataset generation session and then a training session.

I take an existing net, and then using my own unique dataset of sfens which I created and using training parameters which I set ( which is more or less a guessing game ), I run multiple training epochs ( one epoch is one training cycle where tens to hundreds of millions of sfens are compared to a million or so validation sfens. The goal is have to have successful epochs. When you have training session that goes to 20 or so successful epochs, that is a great training run ( each epoch is either accepted, meaning the training session found better weights (success), or rejected, meaning that no better weights were found during the training session (failed)).

The secret to the recipe is how good is your dataset and how good are your training parameters. You really don't know what you have until you starting using your newly created net in a game.

[Dann Corbit]

This post is genius.
The same thing is true (for instance) for things like the Calculus. Newton and Leibniz did not invent Calculus. They merely discovered it.