Page 4 of 14
Re: BrainLearn
Posted: Sun Jan 26, 2020 11:16 pm
by Eduard
8/2Q5/p4kr1/2p5/P3P1p1/2P1R3/3q2P1/6K1 w - - 0 1
Analysis by ShashChess 10 64 POPCNT:
42.Dxc5 Dc1+ 43.Kh2 g3+ 44.Txg3 Dh6+ 45.Th3 Df4+ 46.Kh1 Dc1+ 47.Dg1 Dxg1+ 48.Kxg1 Ke5 49.Te3 Tf6 50.g3 Th6 51.Kf2 Tg6 52.Kg2 Tg8 53.Kf3 Tf8+ 54.Ke2 Tg8 55.c4 Tc8 56.Kd3 Td8+ 57.Kc3 Tc8 58.Tf3 Kxe4 59.Tf6 a5 60.g4 Td8 61.c5 Ke5 62.Tf5+ Ke6 63.Kc4 Td1 64.c6 Tg1
+/= (0.70) Tiefe: 31/64 00:01:04 195MN, tb=769301
Re: BrainLearn
Posted: Mon Jan 27, 2020 8:06 am
by marsell
Eduard wrote: ↑Sun Jan 26, 2020 11:02 pm
I also play on PlayChess (not yet with Brainlearn), and have also seen similar Brainlearn evals. A BUG?
An old problem which has not been solved and is ignored. Probably the real testers are blind. Best you ask Andrea yourself!
Re: BrainLearn
Posted: Tue Jan 28, 2020 10:08 am
by Eduard
Shashchess and new SugarNN have the same BUG. SugarNN now also uses the Shashchess styles. An error has been made here.
Here is a analysis from my last losed game today with Brainlearn 7:
[d]1r2brk1/p5p1/1qn2pPp/2b1pP1Q/1pppP2P/P2P1R2/1PP3BN/1RB3K1 b - - 0 1
Analysis by BrainLearn 7 64 POPCNT:
27...cxd3 28.Lf1 Tf7 29.Sg4 Kf8 30.Lxd3 Ke7 31.Lxh6 gxh6 32.Dxh6 Tf8 33.g7 Ld7 34.Dh7 Kd6 35.gxf8D+ Txf8 36.axb4 Sxb4 37.h5 Dd8 38.Tg3 De7 39.Dxe7+ Kxe7 40.Kg2 Tg8 41.h6 Th8 42.Th3
+/= (0.67 --) Tiefe: 27/38 00:01:56 274MN
Analysis by Stockfish 11 64 POPCNT:
27...b3 28.cxb3 Dxb3 29.Lxh6 gxh6 30.Dxh6 Db7 31.Tg3 Dg7 32.Dc1 c3 33.b4 Ld6 34.h5 a5 35.b5 Sd8 36.a4 Kh8 37.h6 Dc7 38.Lf3 Sf7 39.g7+ Kh7 40.Lh5 Sxh6 41.gxf8S+ Lxf8 42.Lxe8 Txe8 43.b6 Df7 44.Tg6 Lg7 45.Dd1 Kh8
+- (4.25 --) Tiefe: 26/41 00:01:24 205MN
Re: BrainLearn
Posted: Thu Jan 30, 2020 2:17 pm
by Zerbinati
SugaR has been corrected in the latest release:
https://github.com/Zerbinati/SugaR-NN/r ... tag/260120
[d]1r2brk1/p5p1/1qn2pPp/2b1pP1Q/1pppP2P/P2P1R2/1PP3BN/1RB3K1 b - - 0 1
Analysis by SugaR-NN 260120 64 BMI2:
1...b3 2.cxb3 Dxb3 3.Tg3 Da2 4.Axh6 Dxb1+ 5.Cf1 Tb7 6.Ac1 Tff7 7.Dh7+ Rf8 8.Dh8+ Re7 9.gxf7 Axf7 10.Dc8 Cd8 11.Dxc5+ Re8 12.Ad2 Txb2 13.Aa5 Tb5 14.Dxa7 Txa5 15.Dxa5 c3 16.Txg7 Dxd3 17.Dc7 Dc4 18.Dxc4 Axc4 19.Tc7 Ab3 20.Cg3 Cf7 21.Af1 Rd8 22.Tc6 Aa4 23.Tc5 Ch6 24.Rf2 Cg4+ 25.Rf3 Ch2+ 26.Rg2 Cxf1 27.Rxf1 Re7 28.Re2 Ac2 29.h5
Il Bianco sta chiaramente vincendo: +- (7.73 --) Profondità: 29/66 00:01:18 3425MN, tb=86
Re: BrainLearn
Posted: Thu Jan 30, 2020 3:09 pm
by Eduard
Very good. Thank You!
Re: BrainLearn
Posted: Fri Jan 31, 2020 6:36 pm
by Leo
Do you think it will continue to learn after 1,000 games? I have trained it for about 1,000 games so far.
Re: BrainLearn
Posted: Mon Feb 03, 2020 11:54 pm
by amchess
Leo wrote: ↑Fri Jan 31, 2020 6:36 pm
Do you think it will continue to learn after 1,000 games? I have trained it for about 1,000 games so far.
At what time control?
Andrea
Re: BrainLearn
Posted: Tue Feb 04, 2020 3:21 am
by Leo
amchess wrote: ↑Mon Feb 03, 2020 11:54 pm
Leo wrote: ↑Fri Jan 31, 2020 6:36 pm
Do you think it will continue to learn after 1,000 games? I have trained it for about 1,000 games so far.
At what time control?
Andrea
Bullet.
Re: BrainLearn
Posted: Tue Feb 04, 2020 1:36 pm
by amchess
As our tests demonstrate, with 1000 games already there is an improvement,
I didn't do more than 1000 games, but, at a certain pointm there is a bottleneck.
Andrea
Re: BrainLearn
Posted: Tue Feb 04, 2020 5:26 pm
by Leo
amchess wrote: ↑Tue Feb 04, 2020 1:36 pm
As our tests demonstrate, with 1000 games already there is an improvement,
I didn't do more than 1000 games, but, at a certain pointm there is a bottleneck.
Andrea
OK. Thanks.