OKE - one year after

[xr_a_y]

was going for "standard" but maybe this is a french bias

[Terje]

was going for "standard" but maybe this is a french bias

NNUE is about to become the standard

[xr_a_y]

was going for "standard" but maybe this is a french bias

NNUE is about to become the standard

I won't feed the troll

[Rebel]

I tried the sim-test, but Igel doesn't seem to support the uci:depth command.

It should ... This iscoming from 2.7.0 build:

Code: Select all

Igel 2.7.0 64 BMI2 AVX2 by V. Medvedev, V. Shcherbyna
go depth 10
info depth 1 seldepth 1 score cp 30 time 0 nodes 25 tbhits 0 pv c2c4
info depth 2 seldepth 2 score cp 108 time 16 nodes 67 tbhits 0 pv e2e4 g7g6
info depth 3 seldepth 3 score cp 44 time 16 nodes 147 tbhits 0 pv e2e4 g7g6 h2h3
info depth 4 seldepth 4 score cp 110 time 16 nodes 293 tbhits 0 pv e2e4 e7e6 d2d4 h7h6
info depth 5 seldepth 9 score cp 92 time 16 nodes 719 tbhits 0 pv e2e4 e7e6 d2d4 h7h6 c2c4
info depth 6 seldepth 9 score cp 38 time 16 nodes 1830 tbhits 0 pv e2e4 c7c5 g1f3 g8f6 b1c3 d7d6
info depth 7 seldepth 10 score cp 56 time 16 nodes 4632 tbhits 0 pv d2d4 d7d5 c2c4 d5c4 g1f3 g8f6 b1c3
info depth 8 seldepth 17 score cp 68 time 16 nodes 6593 tbhits 0 pv d2d4 d7d5 c2c4 e7e6 b1c3 g8f6 g1f3 c7c5 c4d5
info depth 9 seldepth 19 score cp 68 time 31 nodes 9375 tbhits 0 pv d2d4 d7d5 c2c4 e7e6 g1f3 g8f6 b1c3 c7c5 c4d5
info depth 10 seldepth 21 score cp 52 time 47 nodes 16198 tbhits 0 pv d2d4 e7e6 c2c4 g8f6 g1f3 f8b4 c1d2 c7c5 d2b4 c5b4 d4d5 e6d5 c4d5

I stand corrected, MEA feeds engines with: go movetime 100 depth 1, perhaps this is the problem, most engines give depth priority over movetime.

I wonder if you have by chance the uci output of Igel when you run sim test?

I am currently running the simtest with Lc0, Sf12, Minic, Igel and Rubichess at 100ms, 1000ms and 2500ms. Will post when finished.

[Rebel]

Added Rubichess, ran it with the Igel NNUE.

http://rebel13.nl/html/2-moves.html
http://rebel13.nl/html/3-moves.html
http://rebel13.nl/html/4-moves.html
http://rebel13.nl/html/5-moves.html
http://rebel13.nl/html/6-moves.html

Crazy

Can you also test "regular" Rubi-1.8? Would be interesting to see how good (bad) it is compared to the other non-NNUE engines.

Regards, Andreas

Will do.

[Rebel]

NNUE sim results

http://rebel13.nl/html/mysim.html [100 ms]
http://rebel13.nl/html/mysim-1000ms.html
http://rebel13.nl/html/mysim-2500ms.html

Note that engines like Lc0 and SF12 at 1 second per move already perform at ~3200 elo and so it's not a big suprise they are finding more and more similar moves with increasing time controls. The result is better than I expected.

[xr_a_y]

What is "Rubichess NNUE_Minic".
Is that Rubi with the same SV net as Minic ?

Please, can you give credits to nets authors by using a naming convention like ?

ENGINE_NNUE_NETVERSION

For instance

Minic_NNUE_SV2257

[Laskos]

NNUE sim results

http://rebel13.nl/html/mysim.html [100 ms]
http://rebel13.nl/html/mysim-1000ms.html
http://rebel13.nl/html/mysim-2500ms.html

Note that engines like Lc0 and SF12 at 1 second per move already perform at ~3200 elo and so it's not a big suprise they are finding more and more similar moves with increasing time controls. The result is better than I expected.

Great!

My opinion about the convergence on Sim I observed too between SF12 and Lc0 at longer and longer TC is that in mostly quiet positions like Sim ones, Lc0 is incredibly strong and SF12 approaches it with longer and longer TC. One can check that by doing the Sim of Lc0 at 100ms vs Lc0 2500ms (above 70%) and SF12 at 100ms vs 2500ms (below 60%).
Still not sure about the meaning of the convergence. Do they converge to play "perfect" positional chess at longer TC? Maybe they have the same move ordering, that's why they converge? And even the perfect chess often probably has multiple perfect moves in quiet positions, so there is not necessarily a convergence even if they play perfect moves.

[Rebel]

NNUE sim results

http://rebel13.nl/html/mysim.html [100 ms]
http://rebel13.nl/html/mysim-1000ms.html
http://rebel13.nl/html/mysim-2500ms.html

Note that engines like Lc0 and SF12 at 1 second per move already perform at ~3200 elo and so it's not a big suprise they are finding more and more similar moves with increasing time controls. The result is better than I expected.

And here are the results of the normal AB editions. Same pattern but NNUE nets add 5-6% extra similarity of which a couple of percents are due to the increased playing strength.

http://rebel13.nl/html/ab-100ms.html
http://rebel13.nl/html/ab-1000ms.html
http://rebel13.nl/html/ab-2500ms.html

[Rebel]

NNUE sim results

http://rebel13.nl/html/mysim.html [100 ms]
http://rebel13.nl/html/mysim-1000ms.html
http://rebel13.nl/html/mysim-2500ms.html

Note that engines like Lc0 and SF12 at 1 second per move already perform at ~3200 elo and so it's not a big suprise they are finding more and more similar moves with increasing time controls. The result is better than I expected.

Great!

My opinion about the convergence on Sim I observed too between SF12 and Lc0 at longer and longer TC is that in mostly quiet positions like Sim ones, Lc0 is incredibly strong and SF12 approaches it with longer and longer TC. One can check that by doing the Sim of Lc0 at 100ms vs Lc0 2500ms (above 70%) and SF12 at 100ms vs 2500ms (below 60%).
Still not sure about the meaning of the convergence. Do they converge to play "perfect" positional chess at longer TC? Maybe they have the same move ordering, that's why they converge? And even the perfect chess often probably has multiple perfect moves in quiet positions, so there is not necessarily a convergence even if they play perfect moves.

I think it's the same relationship with the increase of draws with increased time control.