Update on Rebel-15-with-ChrisW-NNUE dev-v2.2

Rebel-15-with-ChrisW-NNUE dev-v2.2, bit of a mouthful, but that's its name at the moment.

First off, it works. I've been playing test games, so far against the Toga4-with-prior-NNUE, and keeping tabs on progress.

The new network inference engine is faster, and able to handle all the prior nets. Tests with the new inference engine vs the old inference engine, both inside Toga4, showed a 6%, 7% increase in nps (old version was getting 0.91 Mnps on my test machine, the new getting 0.99 Mnps). They ran for a few thousand test games at 40/10, W=935, L=807, D=2153 =52%.
Then got the new larger net architecture going, I don't have a net from Ed yet, so tested with a few of my own nets. Nps is down, of course, net is twice the size, at 0.7 Mnps. Tried with three nets, 40/10, results varied between 54% and 56%. My other usual test is to grab a few hundred of the first PGNS, load them into Scid and count the relative wins and losses before move 50. Well, the batch I looked at, had something like 25 wins for the new net before it lost one game, which was about move 60, I think. That's a lot of Tal for a 56% overall run. Not sure where that comes from, no doubt Ed will discover it, but that's pretty good.
Then tried a net at 40/40, which is running now, showing 57% with the early win rate down a little.
I'ld imagine that an Ed-trained net with Ed-tuning tweaks will be able to improve on that, so I'll send him the code in the morning and over to Ed.

They both like attacking ...

[Event "?"]
[Site "?"]
[Date "2022.03.15"]
[Round "388"]
[White "Toga-NNUE-v1.1"]
[Black "Rebel-15-with-ChrisW-NNUE-v2.2"]
[Result "1-0"]
[ECO "B23"]
[GameDuration "00:00:49"]
[GameEndTime "2022-03-15T23:43:49.044 Romance Standard Time"]
[GameStartTime "2022-03-15T23:42:59.315 Romance Standard Time"]
[Opening "Sicilian"]
[PlyCount "67"]
[TimeControl "40/40"]
[Variation "Closed"]

1.e4
{book}
1...c5
{book}
2.Nc3
{book}
2...e6
{book}
3.Nf3
{book}
3...a6
{book}
4.d4
{book}
4...cxd4
{book}
5.Nxd4
{book}
5...Qc7
{book}
6.Bd3
{book}
6...Nf6
{book}
7.Be3
{book}
7...Bd6
{book}
8.f3
{+0.18/12 1.1s}
8...Bb4
{-0.20/11 1.1s}
9.Ne2
{0.00/14 1.2s}
9...Nc6
{-0.04/12 1.5s}
10.O-O
{+0.07/13 0.78s}
10...O-O
{-0.12/13 1.3s}
11.Kh1
{+0.08/13 1.2s}
11...Ba5
{-0.09/14 1.1s}
12.Bg5
{+0.30/13 1.1s}
12...Ne8
{-0.07/12 0.71s}
13.a4
{+0.44/15 1.1s}
13...d6
{-0.11/14 1.8s}
14.f4
{+0.58/15 0.97s}
14...Bd7
{-0.11/13 0.83s}
15.f5
{+0.49/14 1.2s}
15...Ne5
{-0.26/12 1.3s}
16.Be7
{+1.16/14 0.77s}
16...Nf6
{-0.36/12 0.89s}
17.Bxf6
{+1.10/12 1.2s}
17...gxf6
{-0.58/12 0.88s}
18.Ng3
{+0.83/13 0.92s}
18...Kh8
{-0.61/13 0.99s}
19.Nce2
{+1.29/14 1.7s}
19...Bb6
{-0.18/12 1.2s}
20.Nf4
{+1.12/14 1.2s}
20...Qd8
{-0.52/12 1.8s}
21.Be2
{+1.04/12 0.91s}
21...Qe7
{-0.43/13 1.2s}
22.Qd2
{+1.45/13 0.88s}
22...Rg8
{-0.55/15 1.1s}
23.Nfh5
{+1.64/14 0.77s}
23...Raf8
{-0.71/15 0.93s}
24.b4
{+1.62/12 1.3s}
24...Bc7
{-0.81/13 0.88s}
25.b5
{+2.05/13 1.1s}
25...Be8
{-0.78/11 0.70s}
26.Qh6
{+1.79/12 1.7s}
26...Bd8
{-0.82/13 0.74s}
27.Rf4
{+2.31/12 0.81s}
27...Nd7
{-1.11/11 0.81s}
28.Rh4
{+3.26/15 1.2s}
28...Ba5
{-1.63/11 0.81s}
29.Rf1
{+4.38/15 0.97s}
29...Bc3
{-1.66/12 0.83s}
30.e5
{+6.52/13 1.0s}
30...dxe5
{-M8/14 0.80s}
31.Qxh7+
{+M7/34 0.27s}
31...Kxh7
{-M6/4 0.002s}
32.Nxf6+
{+M5/63 0.028s}
32...Kg7
{-M4/4 0.001s}
33.Rh7+
{+M3/63 0.019s}
33...Kxf6
{-M2/4 0.001s}
34.Ne4#
{+M1/63 0.026s, White mates}
1-0



[Event "?"]
[Site "?"]
[Date "2022.03.16"]
[Round "1738"]
[White "Toga-NNUE-v1.1"]
[Black "Rebel-15-with-ChrisW-NNUE-v2.2"]
[Result "0-1"]
[ECO "A09"]
[GameDuration "00:00:44"]
[GameEndTime "2022-03-16T00:53:04.448 Romance Standard Time"]
[GameStartTime "2022-03-16T00:52:19.456 Romance Standard Time"]
[Opening "Reti"]
[PlyCount "60"]
[TimeControl "40/40"]
[Variation "Advance Variation"]

1.Nf3
{book}
1...d5
{book}
2.c4
{book}
2...d4
{book}
3.e3
{book}
3...c5
{book}
4.b4
{book}
4...dxe3
{book}
5.fxe3
{book}
5...cxb4
{book}
6.d4
{book}
6...Nf6
{book}
7.a3
{book}
7...bxa3
{book}
8.Bd3
{+0.02/12 1.1s}
8...Nc6
{-0.05/11 0.79s}
9.d5
{+0.09/12 0.94s}
9...e5
{-0.27/11 0.92s}
10.dxc6
{+0.12/12 1.4s}
10...e4
{-0.17/11 1.1s}
11.Qa4
{+0.37/12 0.69s}
11...b5
{-0.03/10 0.71s}
12.cxb5
{+0.67/11 0.70s}
12...exf3
{-0.53/8 0.81s}
13.Be4
{+0.38/11 1.1s}
13...Nxe4
{-0.36/9 0.73s}
14.Qxe4+
{0.00/11 0.26s}
14...Be6
{-0.16/13 1.1s}
15.O-O
{0.00/13 1.3s}
15...fxg2
{-0.19/12 0.87s}
16.Re1
{+0.21/14 1.2s}
16...Qa5
{-0.14/13 1.4s}
17.Rd1
{-0.56/13 0.94s}
17...Rd8
{0.00/12 1.2s}
18.Bd2
{-1.07/14 1.2s}
18...Be7
{+0.40/12 1.5s}
19.Qe5
{-1.00/13 1.2s}
19...Qb6
{+0.77/14 1.9s}
20.Qxg7
{-1.54/13 0.82s}
20...Rf8
{+0.81/11 0.38s}
21.Qg3
{-2.11/13 0.92s}
21...Bd6
{+1.02/11 1.2s}
22.Qg5
{-1.91/14 1.5s}
22...h6
{+1.34/12 1.3s}
23.Qh4
{-3.02/12 2.0s}
23...a2
{+1.89/11 0.74s}
24.Nc3
{-3.40/11 1.6s}
24...Be5
{+1.70/10 0.51s}
25.Qe4
{-3.50/11 1.2s}
25...Qc5
{+2.43/10 0.58s}
26.Rxa2
{-3.67/11 2.5s}
26...Rg8
{+2.76/10 0.41s}
27.Qxe5
{-4.17/8 1.7s}
27...Qxe5
{+4.54/11 0.37s}
28.Rxa7
{-4.30/8 0.37s}
28...Rxd2
{+7.70/16 0.96s}
29.Re1
{-M4/16 0.85s}
29...Qxh2+
{+M3/63 0.029s}
30.Kxh2
{-M2/4 0.001s}
30...g1=Q#
{+M1/63 0.025s, Black mates}
0-1

[pgn]Event "?"]
[Site "?"]
[Date "2022.03.15"]
[Round "388"]
[White "Toga-NNUE-v1.1"]
[Black "Rebel-15-with-ChrisW-NNUE-v2.2"]
[Result "1-0"]
[ECO "B23"]
[GameDuration "00:00:49"]
[GameEndTime "2022-03-15T23:43:49.044 Romance Standard Time"]
[GameStartTime "2022-03-15T23:42:59.315 Romance Standard Time"]
[Opening "Sicilian"]
[PlyCount "67"]
[TimeControl "40/40"]
[Variation "Closed"]

1.e4
{book}
1...c5
{book}
2.Nc3
{book}
2...e6
{book}
3.Nf3
{book}
3...a6
{book}
4.d4
{book}
4...cxd4
{book}
5.Nxd4
{book}
5...Qc7
{book}
6.Bd3
{book}
6...Nf6
{book}
7.Be3
{book}
7...Bd6
{book}
8.f3
{+0.18/12 1.1s}
8...Bb4
{-0.20/11 1.1s}
9.Ne2
{0.00/14 1.2s}
9...Nc6
{-0.04/12 1.5s}
10.O-O
{+0.07/13 0.78s}
10...O-O
{-0.12/13 1.3s}
11.Kh1
{+0.08/13 1.2s}
11...Ba5
{-0.09/14 1.1s}
12.Bg5
{+0.30/13 1.1s}
12...Ne8
{-0.07/12 0.71s}
13.a4
{+0.44/15 1.1s}
13...d6
{-0.11/14 1.8s}
14.f4
{+0.58/15 0.97s}
14...Bd7
{-0.11/13 0.83s}
15.f5
{+0.49/14 1.2s}
15...Ne5
{-0.26/12 1.3s}
16.Be7
{+1.16/14 0.77s}
16...Nf6
{-0.36/12 0.89s}
17.Bxf6
{+1.10/12 1.2s}
17...gxf6
{-0.58/12 0.88s}
18.Ng3
{+0.83/13 0.92s}
18...Kh8
{-0.61/13 0.99s}
19.Nce2
{+1.29/14 1.7s}
19...Bb6
{-0.18/12 1.2s}
20.Nf4
{+1.12/14 1.2s}
20...Qd8
{-0.52/12 1.8s}
21.Be2
{+1.04/12 0.91s}
21...Qe7
{-0.43/13 1.2s}
22.Qd2
{+1.45/13 0.88s}
22...Rg8
{-0.55/15 1.1s}
23.Nfh5
{+1.64/14 0.77s}
23...Raf8
{-0.71/15 0.93s}
24.b4
{+1.62/12 1.3s}
24...Bc7
{-0.81/13 0.88s}
25.b5
{+2.05/13 1.1s}
25...Be8
{-0.78/11 0.70s}
26.Qh6
{+1.79/12 1.7s}
26...Bd8
{-0.82/13 0.74s}
27.Rf4
{+2.31/12 0.81s}
27...Nd7
{-1.11/11 0.81s}
28.Rh4
{+3.26/15 1.2s}
28...Ba5
{-1.63/11 0.81s}
29.Rf1
{+4.38/15 0.97s}
29...Bc3
{-1.66/12 0.83s}
30.e5
{+6.52/13 1.0s}
30...dxe5
{-M8/14 0.80s}
31.Qxh7+
{+M7/34 0.27s}
31...Kxh7
{-M6/4 0.002s}
32.Nxf6+
{+M5/63 0.028s}
32...Kg7
{-M4/4 0.001s}
33.Rh7+
{+M3/63 0.019s}
33...Kxf6
{-M2/4 0.001s}
34.Ne4#
{+M1/63 0.026s, White mates}
1-0[/pgn]

[pgn][Event "?"]
[Site "?"]
[Date "2022.03.16"]
[Round "1738"]
[White "Toga-NNUE-v1.1"]
[Black "Rebel-15-with-ChrisW-NNUE-v2.2"]
[Result "0-1"]
[ECO "A09"]
[GameDuration "00:00:44"]
[GameEndTime "2022-03-16T00:53:04.448 Romance Standard Time"]
[GameStartTime "2022-03-16T00:52:19.456 Romance Standard Time"]
[Opening "Reti"]
[PlyCount "60"]
[TimeControl "40/40"]
[Variation "Advance Variation"]

1.Nf3
{book}
1...d5
{book}
2.c4
{book}
2...d4
{book}
3.e3
{book}
3...c5
{book}
4.b4
{book}
4...dxe3
{book}
5.fxe3
{book}
5...cxb4
{book}
6.d4
{book}
6...Nf6
{book}
7.a3
{book}
7...bxa3
{book}
8.Bd3
{+0.02/12 1.1s}
8...Nc6
{-0.05/11 0.79s}
9.d5
{+0.09/12 0.94s}
9...e5
{-0.27/11 0.92s}
10.dxc6
{+0.12/12 1.4s}
10...e4
{-0.17/11 1.1s}
11.Qa4
{+0.37/12 0.69s}
11...b5
{-0.03/10 0.71s}
12.cxb5
{+0.67/11 0.70s}
12...exf3
{-0.53/8 0.81s}
13.Be4
{+0.38/11 1.1s}
13...Nxe4
{-0.36/9 0.73s}
14.Qxe4+
{0.00/11 0.26s}
14...Be6
{-0.16/13 1.1s}
15.O-O
{0.00/13 1.3s}
15...fxg2
{-0.19/12 0.87s}
16.Re1
{+0.21/14 1.2s}
16...Qa5
{-0.14/13 1.4s}
17.Rd1
{-0.56/13 0.94s}
17...Rd8
{0.00/12 1.2s}
18.Bd2
{-1.07/14 1.2s}
18...Be7
{+0.40/12 1.5s}
19.Qe5
{-1.00/13 1.2s}
19...Qb6
{+0.77/14 1.9s}
20.Qxg7
{-1.54/13 0.82s}
20...Rf8
{+0.81/11 0.38s}
21.Qg3
{-2.11/13 0.92s}
21...Bd6
{+1.02/11 1.2s}
22.Qg5
{-1.91/14 1.5s}
22...h6
{+1.34/12 1.3s}
23.Qh4
{-3.02/12 2.0s}
23...a2
{+1.89/11 0.74s}
24.Nc3
{-3.40/11 1.6s}
24...Be5
{+1.70/10 0.51s}
25.Qe4
{-3.50/11 1.2s}
25...Qc5
{+2.43/10 0.58s}
26.Rxa2
{-3.67/11 2.5s}
26...Rg8
{+2.76/10 0.41s}
27.Qxe5
{-4.17/8 1.7s}
27...Qxe5
{+4.54/11 0.37s}
28.Rxa7
{-4.30/8 0.37s}
28...Rxd2
{+7.70/16 0.96s}
29.Re1
{-M4/16 0.85s}
29...Qxh2+
{+M3/63 0.029s}
30.Kxh2
{-M2/4 0.001s}
30...g1=Q#
{+M1/63 0.025s, Black mates}
0-1
[/pgn]

This is a new Tal ?

Mclane wrote:

This is a new Tal ?

No. It’s an intermediate stage towards producing a releasable Rebel-15. Both Ed and I train our own nets and write our own search and there are no plans for releasing an Ed search with Chris net, nor a Chris search with an Ed net.

That said, I think Ed is getting some good results already and some good games. I guess he might publish games or even run it on GRL as a private engine, dunno, all up to him. My part would only be fixing net bugs or maybe suggesting some more net enhancements.

Mclane wrote:

This is a new Tal ?

Nope, what Chris said.

Chris created new net structures which (for instance) allows me to upgrade the existing Rebel 14.2 net and gain elo. We exchange nets for the benefit of both as in this case.

Admin wrote:

[pgn][Event "?"]
[Site "?"]
[Date "2022.03.16"]
[Round "1738"]
[White "Toga-NNUE-v1.1"]
[Black "Rebel-15-with-ChrisW-NNUE-v2.2"]
[Result "0-1"]
[ECO "A09"]
[GameDuration "00"]
[GameEndTime "2022-03-16T00.448 Romance Standard Time"]
[GameStartTime "2022-03-16T00.456 Romance Standard Time"]
[Opening "Reti"]
[PlyCount "60"]
[TimeControl "40/40"]
[Variation "Advance Variation"]

1.Nf3
{book}
1...d5
{book}
2.c4
{book}
2...d4
{book}
3.e3
{book}
3...c5
{book}
4.b4
{book}
4...dxe3
{book}
5.fxe3
{book}
5...cxb4
{book}
6.d4
{book}
6...Nf6
{book}
7.a3
{book}
7...bxa3
{book}
8.Bd3
{+0.02/12 1.1s}
8...Nc6
{-0.05/11 0.79s}
9.d5
{+0.09/12 0.94s}
9...e5
{-0.27/11 0.92s}
10.dxc6
{+0.12/12 1.4s}
10...e4
{-0.17/11 1.1s}
11.Qa4
{+0.37/12 0.69s}
11...b5
{-0.03/10 0.71s}
12.cxb5
{+0.67/11 0.70s}
12...exf3
{-0.53/8 0.81s}
13.Be4
{+0.38/11 1.1s}
13...Nxe4
{-0.36/9 0.73s}
14.Qxe4+
{0.00/11 0.26s}
14...Be6
{-0.16/13 1.1s}
15.O-O
{0.00/13 1.3s}
15...fxg2
{-0.19/12 0.87s}
16.Re1
{+0.21/14 1.2s}
16...Qa5
{-0.14/13 1.4s}
17.Rd1
{-0.56/13 0.94s}
17...Rd8
{0.00/12 1.2s}
18.Bd2
{-1.07/14 1.2s}
18...Be7
{+0.40/12 1.5s}
19.Qe5
{-1.00/13 1.2s}
19...Qb6
{+0.77/14 1.9s}
20.Qxg7
{-1.54/13 0.82s}
20...Rf8
{+0.81/11 0.38s}
21.Qg3
{-2.11/13 0.92s}
21...Bd6
{+1.02/11 1.2s}
22.Qg5
{-1.91/14 1.5s}
22...h6
{+1.34/12 1.3s}
23.Qh4
{-3.02/12 2.0s}
23...a2
{+1.89/11 0.74s}
24.Nc3
{-3.40/11 1.6s}
24...Be5
{+1.70/10 0.51s}
25.Qe4
{-3.50/11 1.2s}
25...Qc5
{+2.43/10 0.58s}
26.Rxa2
{-3.67/11 2.5s}
26...Rg8
{+2.76/10 0.41s}
27.Qxe5
{-4.17/8 1.7s}
27...Qxe5
{+4.54/11 0.37s}
28.Rxa7
{-4.30/8 0.37s}
28...Rxd2
{+7.70/16 0.96s}
29.Re1
{-M4/16 0.85s}
29...Qxh2+
{+M3/63 0.029s}
30.Kxh2
{-M2/4 0.001s}
30...g1=Q#
{+M1/63 0.025s, Black mates}
0-1
[/pgn]

This is a high standard of play IMO. It looks simple, but I really struggled to understand several of the moves. This is in contrast to Mclane's 8-bit tournament, where I felt I understood most of the moves easily.

Yes. 8 bit is easy for us. 32 or 64 bit is difficult for our brains.
I once tried to visit chris in france. I was not succesful although i saw his letterbox in front of me.
With me were my son and my father.
3 czubs. Maybe too much.

When will there be a new CSTAL with new gambit book.

Any news?

No news.

Work in progress.

The usual question: are there any notable news?

Working on a new neural net, so far small improvement, too little for release. The obvious way to improve is to add more positions so I am now busy doubling the size from 1.5 billion to 3 billion and that takes a lot of computer time even when you have 72 active threads available for the job.

Chris,

I really look forward to seeing the end product of CSTal. In the mean time I am enjoying viewing the games of the 10 minute version.

[pgn]
[Event "HIARCS 15.0-CSTal match (2 games, 15 minutes per game plus 3 seconds per move)"]
[Site "?"]
[Date "2022.04.06"]
[Round "2"]
[White "CSTal"]
[Black "HIARCS 15.0"]
[Result "1/2-1/2"]
[EventDate "2022.04.06"]
[ECO "E12"]

1.d4 Nf6 2.Nf3 b6 3.c4 Bb7 4.Nc3 e6 5.a3 Ne4 6.Nxe4 Bxe4 7.Nd2 Bg6 8.g3 Nc6 9.e3 e5 10.d5 Na5 11.h4 h6 12.e4 {[%emt 0:00:22] [%eval 54,19]} 12...Bc5 {[%emt 0:00:56] [%eval 9,24]} 13.Bd3 {[%emt 0:00:19] [%eval 77,19]} 13...O-O {[%emt 0:00:31] [%eval 0,22]} 14.g4 {[%emt 0:00:21] [%eval 45,20]} 14...f6 {[%emt 0:00:47] [%eval 50,22]} 15.Bc2 {[%emt 0:00:30] [%eval 38,20]} 15...c6 {[%emt 0:00:23] [%eval 4,21]} 16.b4 {[%emt 0:00:26] [%eval 38,22]} 16...Bd4 {[%emt 0:00:21] [%eval 17,23]} 17.Rb1 {[%emt 0:00:26] [%eval 20,22]} 17...Nb7 {[%emt 0:00:14] [%eval 10,23]} 18.h5 {[%emt 0:00:28] [%eval 17,21]} 18...Bh7 {[%emt 0:00:20] [%eval 6,23]} 19.Nf3 {[%emt 0:00:23] [%eval 27,20]} 19...c5 {[%emt 0:00:21] [%eval 24,22]} 20.O-O {[%emt 0:00:17] [%eval 32,19]} 20...Nd6 {[%emt 0:00:20] [%eval 28,23]} 21.Qe2 {[%emt 0:00:22] [%eval 34,20]} 21...Qc7 {[%emt 0:00:15] [%eval 19,22]} 22.Rd1 {[%emt 0:00:25] [%eval 8,19]} 22...Rfe8 {[%emt 0:01:16] [%eval 16,24]} 23.Bd3 {[%emt 0:00:25] [%eval 33,19]} 23...Nf7 {[%emt 0:00:22] [%eval 26,23]} 24.bxc5 {[%emt 0:00:26] [%eval 46,18]} 24...Bxc5 {[%emt 0:00:22] [%eval 13,24]} 25.Nh4 {[%emt 0:00:17] [%eval 51,18]} 25...d6 {[%emt 0:00:36] [%eval 15,25]} 26.Kh2 {[%emt 0:00:16] [%eval 51,18]} 26...Ng5 {[%emt 0:00:46] [%eval 3,26]} 27.Rf1 {[%emt 0:00:19] [%eval 50,18]} 27...Rac8 {[%emt 0:00:17] [%eval 7,23]} 28.f3 {[%emt 0:00:22] [%eval 51,17]} 28...Qd7 {[%emt 0:00:24] [%eval 14,22]} 29.Bc2 {[%emt 0:00:24] [%eval 51,18]} 29...Bd4 {[%emt 0:00:25] [%eval 20,22]} 30.Rb4 {[%emt 0:00:29] [%eval 64,19]} 30...Rf8 {[%emt 0:00:38] [%eval 21,23]} 31.Ba4 {[%emt 0:00:21] [%eval 72,19]} 31...Qe7 {[%emt 0:00:29] [%eval 31,24]} 32.Bb5 {[%emt 0:00:15] [%eval 97,19]} 32...Rc7 {[%emt 0:00:32] [%eval 24,24]} 33.Kg2 {[%emt 0:00:19] [%eval 99,17]} 33...Qd8 {[%emt 0:00:19] [%eval 32,22]} 34.Ba6 {[%emt 0:00:12] [%eval 97,18]} 34...Bc5 {[%emt 0:00:35] [%eval 35,25]} 35.Rb1 {[%emt 0:00:13] [%eval 97,19]} 35...Qd7 {[%emt 0:00:14] [%eval 38,23]} 36.Rh1 {[%emt 0:00:17] [%eval 101,19]} 36...Bd4 {[%emt 0:00:10] [%eval 40,21]} 37.Re1 {[%emt 0:00:13] [%eval 98,19]} 37...Bc3 {[%emt 0:00:19] [%eval 37,24]} 38.Rd1 {[%emt 0:00:17] [%eval 93,19]} 38...Bd4 {[%emt 0:00:10] [%eval 42,22]} 39.Rh1 {[%emt 0:00:19] [%eval 96,19]} 39...Nf7 {[%emt 0:00:10] [%eval 46,22]} 40.Qd3 {[%emt 0:00:10] [%eval 109,19]} 40...Nd8 {[%emt 0:00:11] [%eval 29,22]} 41.Rd1 {[%emt 0:00:19] [%eval 104,20]} 41...Nb7 {[%emt 0:00:09] [%eval 21,20]} 42.Bxb7 {[%emt 0:00:11] [%eval 100,22]} 42...Rxb7 {[%emt 0:00:05] [%eval 19,22]} 43.Be3 {[%emt 0:00:09] [%eval 84,22]} 43...Bxe3 {[%emt 0:00:08] [%eval 0,23]} 44.Qxe3 {[%emt 0:00:03] [%eval 69,21]} 44...Rc8 {[%emt 0:00:06] [%eval 0,24]} 45.Rdc1 {[%emt 0:00:09] [%eval 67,22]} 45...Rbc7 {[%emt 0:00:05] [%eval 0,23]} 46.Rb4 {[%emt 0:00:14] [%eval 57,22]} 46...Rc5 {[%emt 0:00:06] [%eval 0,24]} 47.a4 {[%emt 0:00:14] [%eval 52,20]} 47...Kh8 {[%emt 0:00:06] [%eval 0,22]} 48.Qe2 {[%emt 0:00:21] [%eval 49,20]} 48...Qd8 {[%emt 0:00:08] [%eval 0,24]} 49.Qf2 {[%emt 0:00:20] [%eval 32,19]} 49...f5 {[%emt 0:00:06] [%eval 0,22]} 50.exf5 {[%emt 0:00:10] [%eval 55,19]} 50...Rxd5 {[%emt 0:00:06] [%eval 0,23]} 51.cxd5 {[%emt 0:00:08] [%eval 110,20]} 51...Rxc1 {[%emt 0:00:06] [%eval 0,25]} 52.Ng6+ {[%emt 0:00:11] [%eval 110,21]} 52...Bxg6 {[%emt 0:00:08] [%eval 0,21]} 53.hxg6 {[%emt 0:00:10] [%eval 137,20]} 53...Qc8 {[%emt 0:00:06] [%eval 0,24]} 54.Rb2 {[%emt 0:00:09] [%eval 143,21]} 54...Rc5 {[%emt 0:00:05] [%eval 0,25]} 55.Qh4 {[%emt 0:00:09] [%eval 146,20]} 55...Qf8 {[%emt 0:00:06] [%eval 0,25]} 56.Rb5 {[%emt 0:00:07] [%eval 167,19]} 56...Kg8 {[%emt 0:00:06] [%eval 0,22]} 57.Kg3 {[%emt 0:00:07] [%eval 187,20]} 57...Rc1 {[%emt 0:00:04] [%eval 0,20]} 58.a5 {[%emt 0:00:06] [%eval 225,20]} 58...Rc5 {[%emt 0:00:04] [%eval 0,23]} 59.Rxc5 {[%emt 0:00:09] [%eval 464,23]} 59...bxc5 {[%emt 0:00:05] [%eval 0,25]} 60.Qh2 {[%emt 0:00:10] [%eval 503,23]} 60...Qb8 {[%emt 0:00:04] [%eval 0,28]} 61.Qe2 {[%emt 0:00:10] [%eval 121,22]} 61...Qb7 {[%emt 0:00:04] [%eval 0,29]} 62.Qd3 {[%emt 0:00:12] [%eval 109,21]} 62...Kf8 {[%emt 0:00:04] [%eval 0,28]} 63.Kh3 {[%emt 0:00:16] [%eval 48,21]} 63...Kg8 {[%emt 0:00:04] [%eval 0,28]} 64.a6 {[%emt 0:00:10] [%eval 12,20]} 64...Qb8 {[%emt 0:00:04] [%eval 0,28]} 65.Qe2 {[%emt 0:00:10] [%eval 14,20]} 65...Qb3 {[%emt 0:00:05] [%eval 0,25]} 66.Kg2 {[%emt 0:00:09] [%eval 5,21]} 66...Kf8 {[%emt 0:00:05] [%eval 0,27]} 67.Qd2 {[%emt 0:00:08] [%eval 4,21]} 67...Qb4 {[%emt 0:00:05] [%eval 0,26]} 68.Qc1 {[%emt 0:00:05] [%eval 5,20]} 68...Ke7 {[%emt 0:00:06] [%eval 0,21]} 69.g5 {[%emt 0:00:08] [%eval 16,21]} 69...Qf4 {[%emt 0:00:04] [%eval 0,23]} 70.f6+ {[%emt 0:00:06] [%eval 0,22]} 70...Kf8 {[%emt 0:00:05] [%eval 0,25]} 71.Qb1 {[%emt 0:00:07] [%eval 0,23]} 71...Qxg5+ {[%emt 0:00:05] [%eval 0,27]} 72.Kh3 {[%emt 0:00:07] [%eval 0,25]} 72...gxf6 {[%emt 0:00:05] [%eval 0,27]} 73.Qb8+ {[%emt 0:00:06] [%eval 0,24]} 73...Kg7 {[%emt 0:00:04] [%eval 0,28]} 74.Qxa7+ {[%emt 0:00:07] [%eval 0,25]} 74...Kxg6 {[%emt 0:00:03] [%eval 0,27]} 75.Qd7 {[%emt 0:00:06] [%eval 0,25]} 75...Qc1 {[%emt 0:00:04] [%eval 0,27]} 76.a7 {[%emt 0:00:08] [%eval 0,24]} 76...Qh1+ {[%emt 0:00:04] [%eval 0,32]} 77.Kg3 {[%emt 0:00:04] [%eval 0,27]} 77...Qe1+ {[%emt 0:00:04] [%eval 0,30]} 78.Kh3 {[%emt 0:00:07] [%eval 0,28]} 78...Qf1+ {[%emt 0:00:04] [%eval 0,32]} 79.Kh4 {[%emt 0:00:06] [%eval 0,27]} 79...Qf2+ {[%emt 0:00:04] [%eval 0,28]} 80.Kh3 {[%emt 0:00:07] [%eval 0,27]} 80...Qxf3+ {[%emt 0:00:03] [%eval 0,28]} 81.Kh2 {[%emt 0:00:07] [%eval 0,27]} 81...Qf4+ {[%emt 0:00:04] [%eval 0,28]} 82.Kh3 {[%emt 0:00:07] [%eval 0,28]} 82...Qe3+ {[%emt 0:00:04] [%eval 0,28]} 83.Kh4 {[%emt 0:00:06] [%eval 0,28]} 83...Qf2+ {[%emt 0:00:04] [%eval 0,30]} 84.Kh3 {[%emt 0:00:04] [%eval 0,29]} 84...Qf3+ {[%emt 0:00:04] [%eval 0,31]} 85.Kh2 {[%emt 0:00:00] [%eval 0,63]} 85...Qf4+ {[%emt 0:00:04] [%eval 0,30]} 86.Kh3 {[%emt 0:00:05] [%eval 0,29]} 86...Qf3+ {[%emt 0:00:25] [%eval 0,37]} 1/2-1/2
[/pgn]

In general, the main handicap of the CSTal/Rebel versions so far is the search. There is something that Marvin's search does well when paired with an aggressive evaluation function: somehow to find tactics. Yesterday I was analyzing a Sicilian position and while Toga Rebel gave a huge advantage to white, it did not see a mate in six (I can't remember how much time I spent on the position), something Marvin Rebel found immediately. I then tried CSTal 10mn, same problem.

I don't have a clue how the search/eval work together, but time after time I noticed the same patterns: Rebel/CSTal would sacrifice for initiative, and then miss a winning attack; or, in defense, simply succumb to a frontal attack that could have been prevented. In other words, there is no awareness of danger.

I already wrote that CSTal was a strange mixture of Karpov and Kasparov -- prevention and aggressiveness. In complicated opening positions it plays probably better than Toga Rebel. It seems to better understand tempi (e.g., it won't play Be3 before exf4) and space.

matejst wrote:

In general, the main handicap of the CSTal/Rebel versions so far is the search. There is something that Marvin's search does well when paired with an aggressive evaluation function: somehow to find tactics. Yesterday I was analyzing a Sicilian position and while Toga Rebel gave a huge advantage to white, it did not see a mate in six (I can't remember how much time I spent on the position), something Marvin Rebel found immediately. I then tried CSTal 10mn, same problem.

I don't have a clue how the search/eval work together, but time after time I noticed the same patterns: Rebel/CSTal would sacrifice for initiative, and then miss a winning attack; or, in defense, simply succumb to a frontal attack that could have been prevented. In other words, there is no awareness of danger.

I already wrote that CSTal was a strange mixture of Karpov and Kasparov -- prevention and aggressiveness. In complicated opening positions it plays probably better than Toga Rebel. It seems to better understand tempi (e.g., it won't play Be3 before exf4) and space.

It's difficult to comment because we don't know really how the evaluation works either.
There will be 'holes' in NN knowledge where it simply misevaluates, its the same with all nets, deeper search compensates, and the good "understanding" of an NN where there are no holes makes for an overall better system.

Any news? I was hoping for a Rebel 15 with a slightly improved Rebel net (the few glaring problems already discussed here solved), and a similar engine with a mature CSTal net, and some search improvements.

matejst wrote:

Any news? I was hoping for a Rebel 15 with a slightly improved Rebel net (the few glaring problems already discussed here solved), and a similar engine with a mature CSTal net, and some search improvements.

Ed is building a net at the moment, with larger number of training positions and (I think) one the enhancements to the architecture. He is also experimenting with which positions go into the training set and which not. There really is a great deal of experimenting and testing involved, and only so much hardware to do it on. Actually, Ed could do with somebody donating him a PC (doesn’t need to be latest thing) with maybe about 4 cores and a GPU, even a 1060 is enough, believe it or not, for NN training, that’s his bottleneck.

Thanks, Chris.

I have a spare laptop we could use, but it is a AMD A8 with Radeon graphics, I think. So it is not much, but if I can help, I will gladly do it. Just tell me what.

I think I have about 50 elo by now but not happy yet, there must be more. Besides I am not 100% sure if it is still possible to release a SSE version. NNUE eval (by far) works best with AVX2 and new NNUE architectures developed by Chris at some point in the future may no longer work for SSE computers as they become an obstacle for progress.

I hope that Rebel 15 will be SSE compatible, no hard guarantee, but if so there will be also a Rodent NN SSE.

I see I will have to ask money from my old mum and buy a new computer (and hope my children don't kidnap it right away).

matejst wrote:

Thanks, Chris.

I have a spare laptop we could use, but it is a AMD A8 with Radeon graphics, I think. So it is not much, but if I can help, I will gladly do it. Just tell me what.

Laptops are probably not advised for this type of NN training because the training software is so thoroughly optimised (we are using PyTorch) that it continuously hammers both the GPU and CPU and that generates a lot of heat. training can last several days. Heat dissipation in laptops is not terribly good, in general.
There’s quite possibly issues with farming out the training elsewhere, for example, I am currently working with 12 billion training positions and that is targeted to grow plenty more. That’s roughly 500 or 600 Gb of data, it can be compressed and I guess we could change data format and use the SF data format (I think that’s 2-3 bytes per position) but even so that’s ~30Gb for 12B positions. Completely impractical for me on rural French internet and probably not much better in practice with any “home” internet.

50 Elo is a huge increase at this level, this would bring Rebel 15.0 roughly to the level of Fritz 18NN or the new Wasp 5.50. A Rebel 15.0 would also be a nice Easter present. , thanks for your work Ed.

Dio wrote:

50 Elo is a huge increase at this level, this would bring Rebel 15.0 roughly to the level of Fritz 18NN or the new Wasp 5.50. A Rebel 15.0 would also be a nice Easter present. , thanks for your work Ed.

I second this...