Alpha subZero - NN for CPU

[Damir]

190.1 version just released.

https://www.dropbox.com/s/cb26fk2rgh9r7 ... 1.rar?dl=0

[dkappe]

190.1 version just released.

https://www.dropbox.com/s/cb26fk2rgh9r7 ... 1.rar?dl=0

Where is the source? I’ve requested it once before.

[Damir]

Private Engine= no source available.

If author not want to share it it is up to him.

What will you do with the source anyway ?

You have put Thothfish source on github and it has been collecting dust there for months.

If the author post the source of his engine, who is to say we won't see clones of other engines based on his work.

[Joerg Oster]

190.1 version just released.

https://www.dropbox.com/s/cb26fk2rgh9r7 ... 1.rar?dl=0

Where is the source? I’ve requested it once before.

There is definitely something fishy going on.

Look at this output.

Code: Select all

uciok
position startpos moves d2d4 d7d5
go movetime 10000
info depth 5 seldepth 0 multipv 1 score cp 108 nodes 0 nps 374 tbhits 0 time 1 pv e2e3 (none) d2a1r e1a1 f2f3 a1a8
info depth 5 seldepth 1 multipv 1 score cp 41 nodes 9368 nps 20500 tbhits 0 time 457 pv c1f4 c8f5 a7g2 a6a1 a2a4 (none)
info depth 6 seldepth 1 multipv 1 score cp 29 nodes 17298 nps 20472 tbhits 0 time 845 pv c2c3 e7e6 c1f4 g8f6 a1a2 (none) c5g6
info depth 6 seldepth 2 multipv 1 score cp 49 nodes 25192 nps 20316 hashfull 22 tbhits 0 time 1240 pv e2e3 e7e6 f1e2 g8f6 c2c4 f8e7 (none)
info depth 6 seldepth 2 multipv 1 score cp 39 nodes 33032 nps 20277 hashfull 30 tbhits 0 time 1629 pv e2e3 g8f6 g1f3 e7e6 f1e2 f8e7 (none)
info depth 7 seldepth 3 multipv 1 score cp 50 nodes 40715 nps 20216 hashfull 37 tbhits 0 time 2014 pv g1f3 g8f6 c2c4 e7e6 c1g5 b8d7 f1e2 d5c4
info depth 7 seldepth 3 multipv 1 score cp 55 nodes 48468 nps 20186 hashfull 39 tbhits 0 time 2401 pv g1f3 g8f6 c2c4 d5c4 e2e3 b7b5 b2b3 d5c4
info depth 7 seldepth 4 multipv 1 score cp 70 nodes 56231 nps 20169 hashfull 43 tbhits 0 time 2788 pv g1f3 g8f6 c2c4 d5c4 e2e3 b7b5 b2b3 d5c4
info depth 3 currmove g1f3 currmovenumber 1
info depth 3 currmove e2e3 currmovenumber 2
info depth 3 currmove c2c4 currmovenumber 3
info depth 3 currmove e2e4 currmovenumber 4
info depth 3 currmove c1g5 currmovenumber 5
info depth 3 currmove c1f4 currmovenumber 6
info depth 3 currmove b1d2 currmovenumber 7
info depth 3 currmove b1c3 currmovenumber 8
info depth 3 currmove c2c3 currmovenumber 9
info depth 3 currmove f2f4 currmovenumber 10
info depth 3 currmove g2g4 currmovenumber 11
info depth 3 currmove g1h3 currmovenumber 12
info depth 3 currmove c1d2 currmovenumber 13
info depth 3 currmove c1e3 currmovenumber 14
info depth 3 currmove c1h6 currmovenumber 15
info depth 3 currmove d1d2 currmovenumber 16
info depth 3 currmove d1d3 currmovenumber 17
info depth 3 currmove e1d2 currmovenumber 18
info depth 3 currmove b2b3 currmovenumber 19
info depth 3 currmove a2a3 currmovenumber 20
info depth 3 currmove f2f3 currmovenumber 21
info depth 3 currmove g2g3 currmovenumber 22
info depth 3 currmove h2h3 currmovenumber 23
info depth 3 currmove a2a4 currmovenumber 24
info depth 3 currmove b2b4 currmovenumber 25
info depth 3 currmove h2h4 currmovenumber 26
info depth 3 currmove b1a3 currmovenumber 27
info depth 8 seldepth 6 multipv 1 score cp 50 upperbound nodes 867466 nps 19667 hashfull 635 tbhits 0 time 44107 pv g1f3 e7e6 c2c4 d5c4 e2e3 b7b5 b2b3 d5c4 h8c8
info depth 3 currmove g1f3 currmovenumber 1
info depth 8 seldepth 6 multipv 1 score cp 50 nodes 902982 nps 19669 hashfull 654 tbhits 0 time 45908 pv g1f3 e7e6 c2c4 d5c4 e2e3 b7b5 b2b3 d5c4 h8c8
bestmove g1f3 ponder e7e6

The info lines definitely look like those from Stockfish.
The currmove lines are being printed in slow-motion, and there are illegal moves in the given pv.

All UCI options, however, look like they are taken from Lc0.
(But I must admit I'm no expert for Lc0 ...)

[crem]

If the author post the source of his engine, who is to say we won't see clones of other engines based on his work.

Both this thread and (especially) the "subzero" thread on chess2u seems like it's some kind of a game to overtroll an interlocutor?

[peter]

There is definitely something fishy going on.

Look at this output.

Or at the testposition, that was discussed here

http://talkchess.com/forum3/viewtopic.p ... +w#p819396

[d]1rb1qrk1/2b2pp1/p3pBn1/3pP1Pp/1ppP4/2P1QN2/PP3P1P/R2BR1K1 w - - 0 21
21.Nh4!

A most typical one with LC0 much better in solving then SF.

subZero is clueless after 10 minutes at 24 "LC0 Cpu"-threads, 8000 NNCache.
Yet it is about as good at keeping the winning lines in hash in Backward as Fish is.

LC0 and Fat Fritz as shown in the old thread are at the best move almost at once with good hardware, to subbby you have to show a line like this:

21.Nh4 bxc3 (?) 22. bxc3 Nxh4 23. Qh3 Nf5 24. Qxh5 Rb2 25. Kh1 Rxf2 26. Bg4

to make it's eval climb not before 24th move.

After Backward to starting position best first move for White an Black in output:

1rb1qrk1/2b2pp1/p3pBn1/3pP1Pp/1ppP4/2P1QN2/PP3P1P/R2BR1K1 w - -

Engine: alpha subZero 190.1M (0 MB)
von Darin Pann

18/29 2:14 +1.76 1.Sh4 Kh7 2.Lxh5 Th8 3.Dg3 Kg8
4.Lxg6 fxg6 5.Te3 Ld7 6.Tc1 La5
7.Tf3 Th5 8.Sg2 gxf6 9.gxf6 Kf7
10.Sf4 (10.843.937) 80

[Werner]

...even benchmark command works like inside Stockfish:

Position: 41/42
info depth 0 score mate 0
bestmove (none)

Position: 42/42
info depth 2 seldepth 0 multipv 1 score cp 110 nodes 0 nps 243 tbhits 0 time 2 pv d1e3 d2h8 e2e1
info depth 2 seldepth 1 multipv 1 score cp 21 nodes 684 nps 9125 tbhits 0 time 75 pv e7e3 h3c8 a2c5
info depth 3 seldepth 1 multipv 1 score cp 21 nodes 689 nps 9190 tbhits 0 time 75 pv e7e3 h3c8 c2c3 e6a1
info depth 3 seldepth 2 multipv 1 score cp 29 nodes 693 nps 9126 tbhits 0 time 76 pv e7e3 d8e6 c2c3 e6g5
info depth 3 seldepth 2 multipv 1 score cp 22 nodes 698 nps 9069 tbhits 0 time 77 pv e7e3 d8e6 c2c3 e6g5
info depth 4 seldepth 3 multipv 1 score cp 27 nodes 704 nps 9027 tbhits 0 time 78 pv e7e3 d8e6 c2c3 f8e8 e1g2
info depth 4 seldepth 6 multipv 1 score cp 27 nodes 711 nps 9122 tbhits 0 time 78 pv e7e3 d8e6 c2c3 f8e8 e1g2
info depth 4 seldepth 5 multipv 1 score cp 17 nodes 718 nps 9094 tbhits 0 time 79 pv e7e3 d8e6 e3d3 c7c6 d1e3
info depth 5 seldepth 5 multipv 1 score cp 29 nodes 732 nps 9047 tbhits 0 time 81 pv e7e3 b7b6 e1g2 c7c5 c2c3 h3d7
info depth 5 seldepth 8 multipv 1 score cp 5 nodes 783 nps 9112 tbhits 0 time 86 pv e7e3 b7b6 e1g2 f8e8 e3d3 c7c5
info depth 5 seldepth 7 multipv 1 score cp 0 nodes 801 nps 9108 tbhits 0 time 88 pv e7e3 b7b6 e1g2 f8e8 e3d3 c7c5
info depth 6 seldepth 9 multipv 1 score cp 3 nodes 835 nps 9080 tbhits 0 time 92 pv e7e3 b7b6 e1g2 c7c5 c2c3 f8e8 e3d2
info depth 6 seldepth 10 multipv 1 score cp 5 nodes 1065 nps 8957 tbhits 0 time 119 pv e7e3 b7b6 e1g2 c7c5 c2c3 h3d7 b2b
3
bestmove e7e3 ponder b7b6
Benchmark final time 6400 ms, calculating nodes per second 15697.9

[dkappe]

Private Engine= no source available.

If author not want to share it it is up to him.

What will you do with the source anyway ?

You have put Thothfish source on github and it has been collecting dust there for months.

If the author post the source of his engine, who is to say we won't see clones of other engines based on his work.

As stated by the author, it’s based on stockfish and lc0 source, which are both GPLv3. So, fork over the code.

[dkappe]

You have put Thothfish source on github and it has been collecting dust there for months.

So what? Digital assets don’t collect dust. The author only had the source available as a downloadable zip. I simply made it available in a more convenient form. It’s been forked three times, so maybe someone is using it in some way. Great.

Now fork over your code as the GPL requires.

[Damir]

Did you try and ask the author for a source in private mail in a polite way ?

If you openly ask about it and demanding the source released you won't get any I suppose...