Minic version 2

[xr_a_y]

Due the stockfish NNUE hype, i am in the mood to try something with NN !

So I got back to this idea (http://talkchess.com/forum3/viewtopic.p ... na#p828151) of trying to create a model of the error between Minic texel tuned evaluation using as input some eval features (now that I have some thanks to Minic personality thing ...).

For now I'm trying to tune a "simple" 16/32/16/1 MLP based on half of the data of lichess-new-labeled and look at the result on the other half of the data.
The next step will be to fuse Minic eval and this little NN eval and see how it goes...

I'll see where this experiment leads ... probably nowhere ... but at least I'm playing with NN a bit.

[carldaman]

Good luck with your experiments - without experimentation where would we be?

[xr_a_y]

I'm happy to announce Minic 2.43, which seems to be +30 versus previous release for both STC and LTC.

https://github.com/tryingsomestuff/Mini ... s/tag/2.43

[Gregory Owett]

Has anyone had success in downloading minic 2.43 ?

[xr_a_y]

Has anyone had success in downloading minic 2.43 ?

what is your issue ?

[Gregory Owett]

When I download the file "minic 2.43 mingw_x64_skylake.exe", I get a file "damaged or its format is unknown".

[MOBMAT]

Works for me, though I get a "not a common download" warning, but I override it and it continues on.
Seems to work fine.

Code: Select all

FEN: 7k/8/7K/p1P1p1p1/B7/3nN1p1/5r2/6B1 w - - 0 1

Minic 2.43:
 8	00:00	 6k	5,636k	+2.83	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kxg5 Kg7 4.Nf5+ Kf7
 9	00:00	 7k	6,599k	+2.54	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Nf5 g4 4.Kg5 Nc5 5.Bxe5+ Kg8 6.Kxg4
 9	00:00	 7k	7,475k	+2.50	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Bd6+
 10	00:00	 10k	2,456k	+2.48	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Ng4 Nc5 5.Bxe5 a4
 11	00:00	 13k	2,551k	+2.36	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Nc5 5.Nd5 Ne4 6.Nf6+
 12	00:00	 16k	1,604k	+2.14	1.Bxf2 Nxc5 2.Be1 Nxa4 3.Bxg3 Nc5 4.Kg6 a4 5.Bxe5+ Kg8
 12	00:00	 60k	1,763k	+2.19	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Bxe5+ Kg8 4.Kg6 g4 5.Nd5 Kf8 6.Kf5 g3 7.Bd6+
 13	00:00	 65k	1,915k	+2.39	1.Bxf2 Nxc5 2.Bxg3 Kg8 3.Bb5 Ne4
 13	00:00	 72k	2,104k	+2.61	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 Nb6 6.Bd4 Nc8
 14	00:00	 75k	2,197k	+2.37	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Ng4 Nb6 6.Kf5 a4 7.Bd6+ Kf7 8.Ba3 Nd5 9.Ne5+ Kg7
 14	00:00	 163k	2,542k	+2.25	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Bxe5+ Kg8 4.Kg6 Kf8 5.Kf6 g4 6.Bd4 Ke8 7.Ke6 Kd8 8.Ng2 Ke8 9.Kf5 g3
 15	00:00	 165k	2,573k	+2.31	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 g4 6.Bd4 Ke8 7.Ke6 Kd8 8.Kd6
 16	00:00	 187k	2,715k	+2.17	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 g4 6.Bd6+ Ke8 7.Nd5 Kd7 8.Bc7 g3 9.Bxg3
 16	00:00	 298k	3,170k	+1.94	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Ng4 Kf8 5.Bxe5 Nb6 6.Ne3 Ke8 7.Bd6 Na4 8.Kf6 g4 9.Bg3 Kd7 10.Nxg4
 16	00:00	 618k	3,574k	+1.65	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kd5 Nd3 8.Bc3 Nf4+ 9.Kc5 Kf7 10.Ng4
 17	00:00	 1,085k	3,832k	+1.54	1.Bxf2 Nxc5 2.Be1 Nxa4 3.Bxg3 Kg8 4.Kg6 Kf8 5.Nc4 g4 6.Kf5 Nc5 7.Bxe5 Kf7 8.Nd6+ Ke7 9.Nb5 Nd3
 18	00:00	 1,182k	3,849k	+1.55	1.Bxf2 Nxc5 2.Be1 Nxa4 3.Bxg3 Kg8 4.Kg6 Kf8 5.Nc4 e4 6.Kf5 e3 7.Nxe3 Ke7 8.Ke5 Nb6 9.Nf5+ Kf7
 19	00:00	 1,307k	3,900k	+1.47	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Ng4 Kf8 5.Kf5 Nc5 6.Nh6 a4 7.Bxe5 Ke7 8.Bd4 Kd6 9.Be5+ Ke7
 19	00:00	 1,636k	4,049k	+1.42	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf5 Ke7 6.Nc4 Kd7 7.Ke4 Nc5+ 8.Kd5 Nd3 9.Bg3 Ke7 10.Bd6+ Kf7
 20	00:00	 2,368k	4,426k	+1.36	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Kf5 Nc5 5.Bxe5 Kf7 6.Nc4 Nd3 7.Bd4 a4 8.Nd6+ Ke7 9.Nb7
 20	00:00	 3,191k	4,624k	+1.25	1.Bxf2 Nxc5 2.Bxg3
 20	00:00	 4,580k	4,831k	+1.07	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Ng4 Kg8 4.Kg6 Kf8 5.Kf6 Nc5 6.Be1 Ke8 7.Bxa5 e4 8.Kg6 e3 9.Nf6+ Ke7 10.Bb4 Kd6 11.Be1 Nd3 12.Bc3 Kc6 13.Ne8
 20	00:01	 8,792k	5,249k	+0.88	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Nf5 g4 5.Bh4 Kf8 6.Kf6 Nc5 7.Kxe5 Kf7 8.Kd5 Nd7 9.Kd6 Nf6 10.Ke5 Nd7+ 11.Kf4 a4 12.Be7
 21	00:01	 10,403k	5,371k	+0.82	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kd5 Nd3 8.Bd6 Kd7 9.Nf5 Nf4+ 10.Ke4 Kc6 11.Ba3 Kb5 12.Ne3
 21	00:01	 10,408k	5,373k	+0.82	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kd5 Nd3 8.Bd6 Kd7 9.Nf5 Nf4+ 10.Ke4 Kc6 11.Ba3
 22	00:02	 12,483k	5,456k	+0.82	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kd5 Nd3 8.Bd6 Kd7 9.Nf5 Nf4+ 10.Ke4 Kc6 11.Ba3 Kb5
 23	00:02	 12,507k	5,454k	+0.88	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Bxe5+ Kg8 4.Kg6 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kd5 Nd3 8.Bd6 Kd7 9.Nc4 Ke8 10.Nxa5
 23	00:02	 12,544k	5,454k	+0.90	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Bxe5+ Kg8 4.Kg6 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kd5 Nd3 8.Bd6 Kd7 9.Nc4 Nf4+ 10.Ke4 Kc6 11.Ba3 a4
 24	00:02	 12,945k	5,460k	+0.84	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Bxe5+ Kg8 4.Kg6 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kf5 Kd7 8.Bb2 g4 9.Ba3 Kc6 10.Ke5 a4 11.Nf5 Nd3+ 12.Kf6 Ne1 13.Bd6 Nf3 14.Ba3 Kb5
 24	00:03	 17,669k	5,665k	+0.73	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Bxe5+ Kg8 4.Kg6 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kf5 Kd7 8.Ng4 Kc6 9.Bb2 Na4 10.Bf6 Nb6 11.Ne5+ Kb5 12.Bg7 a4 13.Ng6 Nc4
 24	00:04	 25,204k	5,834k	+0.69	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Nf5 Kf8 5.Kf6 Nc5 6.Kxe5 a4 7.Kd5 Nd7 8.Kd6 Nf6 9.Ke6 Ne4 10.Be5 Nc5+ 11.Kd5 Nd3 12.Bd6+ Kf7 13.Ba3 Kf6 14.Ng3 Nf4+
 25	00:04	 26,958k	5,887k	+0.65	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Nf5 Kf8 5.Kf6 Nc5 6.Kxe5 a4 7.Kd5 Nd7 8.Kd6 Ke8 9.Be1 a3 10.Bc3 Nf8 11.Ba1 g4 12.Ng3 Kf7 13.Ne4
 26	00:05	 30,393k	5,954k	+0.59	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kxg5 Nc5 4.Ng2 Kg8 5.Bh2 a4 6.Kf5 a3 7.Kf6 Nd7+ 8.Kg5 Nc5 9.Kf6
 26	00:05	 35,187k	6,058k	+0.54	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Nf5 Kf8 5.Kf6 Nc5 6.Kxe5 a4 7.Kd5 Nd7 8.Kd6 Ke8 9.Be1 a3 10.Bc3 g4 11.Ng3 Nf8 12.Ke5 a2 13.Ba1 Kf7 14.Nf5
 27	00:05	 36,128k	6,068k	+0.54	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Nf5 Kf8 5.Kf6 Nc5 6.Kxe5 a4 7.Kd5 Nd7 8.Kd6 Ke8 9.Be1 a3 10.Bc3 g4 11.Ng3 Nf8 12.Ke5 a2 13.Ba1 Kf7 14.Nf5
 28	00:10	 66,326k	6,145k	+0.48	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 Ke8 6.Bd4 Kd7 7.Kg6 g4 8.Ng2 g3 9.Bb2 Nxb2
 28	00:11	 70,237k	6,191k	+0.37	1.Bxf2 Nxc5 2.Kxg5 gxf2 3.Be8 Ne4+ 4.Kf5 f1Q+ 5.Nxf1 Nd6+ 6.Kxe5
 28	00:13	 85,890k	6,321k	+0.27	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kd6 Nd3 8.Bc3 a4 9.Kd5 Nf4+ 10.Ke4 Ke7 11.Nf5+ Kd7 12.Ba1 Kc6 13.Kd4 a3 14.Ng3 a2 15.Ke5 Nd3+ 16.Ke4 Nc5+
 29	00:15	 99,043k	6,418k	+0.22	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf6 Ke8 6.Ke6 Nc5+ 7.Kd6 Nd3 8.Bc3 a4 9.Kd5 a3 10.Kd4 Nf4 11.Ke4 Kd7 12.Ba1 Ne6 13.Nc4 Nf4 14.Kf5 Kc6 15.Ke5 Nd3+
 30	00:17	 114,698k	6,484k	+0.22	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Ng4 Kf8 5.Kf6 Nc5 6.Be1 Nd3 7.Bc3 a4 8.Kf5 Ke7 9.Ke4 Nc5+ 10.Kxe5 Nd3+ 11.Ke4 Nf4 12.Kf5 Nd3 13.Ne3 a3 14.Nc4 Kf7 15.Bd4 Ke7 16.Bf6+
 31	00:20	 132,871k	6,531k	+0.16	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Bxe5 Kf8 5.Kf5 Ke7 6.Ng4 Kd7 7.Ke4 Nb6 8.Bg7 a4 9.Ne3 Kc6 10.Nf5 a3 11.Ba1 g4 12.Ng3 Kc5 13.Kd3 a2 14.Kc3 Nc4 15.Kb3
 31	00:28	 187,788k	6,519k	+0.05	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Be1 Nc5 5.Bxa5 Nd3 6.Ng4 e4 7.Ne3 g4 8.Kf5 Kf7 9.Ng2 e3 10.Ke4 Nc5+ 11.Kd5 Ne6 12.Ke5 e2 13.Ne1 g3 14.Bb6 g2 15.Bg1 Nf8 16.Kf5 Nd7 17.Be3 Nf6 18.Nxg2
 31	00:30	 198,634k	6,539k	+0.01	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Nf5 Kf8 5.Be1 g4 6.Kf6 Nc5 7.Ng3 Ke8 8.Bxa5 Kd7 9.Kf5 Kc6 10.Bd2 Kd5 11.Be3 Nb7 12.Bg1 Nd6+ 13.Kg6 Ne4 14.Nf5 g3 15.Kh5 Ke6 16.Ng7+ Kf7 17.Nf5 Ke6
 32	00:37	 243,009k	6,473k	 0.00	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Be1 Nc5 5.Bxa5 Nd3 6.Ng4 e4 7.Ne3 g4 8.Kf5 Kf7 9.Ng2 e3 10.Ke4 Nc5+ 11.Kd5 Ne6 12.Ke5 e2 13.Ne1 g3 14.Bb6 g2 15.Bg1 Nf8 16.Kf5 Nd7
 33	00:39	 255,501k	6,471k	 0.00	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Kxg5 Kf7 5.Kf5 Nc5 6.Bf2 a4 7.Nc2 Nd3 8.Be3 a3 9.Nxa3
 34	01:02	 398,087k	6,380k	 0.00	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Kxg5 Kf7 5.Kf5 Nc5 6.Bf2 a4 7.Nc2 Nd3 8.Be3 a3 9.Nxa3
 35	01:07	 426,806k	6,365k	 0.00	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Kxg5 Kf7 5.Kf5 Nc5 6.Bf2 a4 7.Nc2 Nd3 8.Be3 a3 9.Nxa3
 36	01:13	 466,568k	6,371k	 0.00	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Kxg5 Kf7 5.Kf5 Nc5 6.Bf2 a4 7.Nc2 Nd3 8.Be3 a3 9.Nxa3
 37	01:22	 524,087k	6,383k	 0.00	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Kxg5 Kf7 5.Kf5 Nc3 6.Be1 Nb5 7.Ke4 Ke6 8.Kd3 a4 9.Kc2 a3 10.Kb1 a2+ 11.Ka1 Nd6 12.Kxa2
 38	01:23	 531,911k	6,381k	 0.00	1.Bxf2 Nxc5 2.Bxg3 Nxa4 3.Kg6 Kg8 4.Kxg5 Kf7 5.Kf5 Nc3 6.Be1 Nb5 7.Ke4 Ke6 8.Kd3 a4 9.Kc2 a3 10.Kb1 a2+ 11.Ka1 Nd6 12.Kxa2

[xr_a_y]

When I download the file "minic 2.43 mingw_x64_skylake.exe", I get a file "damaged or its format is unknown".

Sorry for the stupid question, but are you sure you are on a 64bits platform (cpu and OS) and with at least a skylake generation CPU ?

[xr_a_y]

Here are some recent results

TC 2min+0.1s
TT 256Mb

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   1 demolito                       68      20     717   59.6%   39.5%
   2 igel-last                      19      19     718   52.8%   46.8%
   3 Vajolet2_2.8                   15      19     718   52.2%   44.0%
   4 minic_2.43                     12      19     717   51.7%   44.9%
   5 texel                          -3      19     717   49.6%   42.0%
   6 minic_2.33                    -23      19     718   46.7%   41.8%
   7 combusken                     -90      21     717   37.4%   34.3%

[xr_a_y]

Minic 2.47 is released. This is a quite special version, as NNUE (from Stockfish) as be integrated inside.
This release has NNUE technology enabled (by #define WITH_NNUE inside definition.hpp), but has to be considerate as "MinicNNUE" only if you specify a network file using the NNUEFile GUI option.

MinicNNUE (using nn-97f742aaefcd.nnue) seems +200Elo versus standard Minic evaluation.
But as already stated in many places, MinicNNUE won't be "official Minic" version because it does not represent at all my own work.
Minic is already of course vastly inspired by others engines but I try and test everything by myself, and those who has followed Minic development know how long it took... but here, integrated NNUE was more or less just a copy/paste.
So please do not "categorized" MinicNNUE with standard Minic (i'm thinking of CCRL, FGRL, CEGT testing here).

Anyone is of course free to use this Minic version with any network file you want (but please call it MinicNNUE, not Minic, as soon as a net is used). I'd also be glad if someone try to train specific net for it...

This experiment showed that many Elo can be gained in Minic by working on its evaluation, and also many more working on search !

Using NNUE inside Minic was quite easy (as discussed on the TCEC discord channel), mainly because the NNUE code itself is very clear, a very good job was done by people who integrated it into Stockfish.

Finally, I'd like to make a point about NNUE technology release. I think many engine dev will try (and success) to use it in the next days or weeks, and it would feel more natural to me if the NNUE is hosted in a specific repository so that many engine authors can work with it. I think this would benefit to everyone in the community, including Stockfish and maybe even shogi engine devs. Maybe some will try to write others architectures for instances. This will at least be better than 100 copy/paste different codes hard to track and merge.