Test position: Neural net blindness.

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

mwyoung
Posts: 2727
Joined: Wed May 12, 2010 10:00 pm

Test position: Neural net blindness.

Post by mwyoung »

[d]r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 33

How long does it take for your NN engine to find Rxe6?



New game Line
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Mephisto Gideon Professional (1993) for Winboard and UCI:

33.Rxe6 fxe6 34.Kb2 Qd4 35.Qxd4 Nd5 36.Qd3 Nxb4 37.Qe4 a5
White is clearly winning: +- (10.20) Depth: 12 00:00:23 61592kN
(, 08.02.2020)
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
Top
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: Test position: Neural net blindness.

Post by zullil »

Not long:

info depth 12 seldepth 43 time 43748 nodes 721516 score cp 808 wdl 982 7 11 hashfull 2 nps 16509 tbhits 11 multipv 1 pv b6e6 d5e6 b1b2 a8c8 c3b3 e6b3 c2b3
Top
mwyoung
Posts: 2727
Joined: Wed May 12, 2010 10:00 pm

Re: Test position: Neural net blindness.

Post by mwyoung »

zullil wrote: ↑Sat Feb 08, 2020 12:21 pm Not long:

info depth 12 seldepth 43 time 43748 nodes 721516 score cp 808 wdl 982 7 11 hashfull 2 nps 16509 tbhits 11 multipv 1 pv b6e6 d5e6 b1b2 a8c8 c3b3 e6b3 c2b3
What NN?

Still 43 seconds is a long time. Even a 1993 engines finds the move instantly. Most NN take a long time.

New game Line
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Lc0 v0.23.2+git.c8d9095 Sergio:

33.Kb2 Qe5 34.Qxe5 Rxe5 35.Rxf6 Ra7 36.Rd6 Re3 37.c3 Rxf3 38.Be2 Rf2 39.Rd2 Rxe2 40.Rxe2 Rd7 41.c4 a5 42.bxa5 bxc4 43.Kc3 Kxa5 44.Kxc4 Ka4 45.Re3 f5 46.h4 Rf7 47.Rf3 Rd7 48.Kc5 Rd2
The position is equal: = (0.03) Depth: 22/43 00:01:01 663kN, tb=56
(, 08.02.2020)

New game Line
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Lc0 v0.23.2+git.c8d9095 Sergio:

33.Kb2 Qe5 34.Qxe5 Rxe5 35.Rxf6 Ra7 36.Rd6 Re3 37.c3 Rxf3 38.Be2 Rf2 39.Rd2 Rxe2 40.Rxe2 Rd7 41.c4 a5 42.bxa5 bxc4 43.Kc3 Kxa5 44.Kxc4 Ka4 45.Re3 Rd2 46.Re7 Rc2+ 47.Kd4 Rxh2 48.Rxf7 Kxa3 49.Ke3 Kb4 50.Kf3 Kc5 51.Rd7 Kc6
The position is equal: = (0.04) Depth: 25/48 00:03:06 2551kN, tb=659
(, 08.02.2020)
Last edited by mwyoung on Sat Feb 08, 2020 12:31 pm, edited 1 time in total.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
Top
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: Test position: Neural net blindness.

Post by zullil »

mwyoung wrote: ↑Sat Feb 08, 2020 12:24 pm
zullil wrote: ↑Sat Feb 08, 2020 12:21 pm Not long:

info depth 12 seldepth 43 time 43748 nodes 721516 score cp 808 wdl 982 7 11 hashfull 2 nps 16509 tbhits 11 multipv 1 pv b6e6 d5e6 b1b2 a8c8 c3b3 e6b3 c2b3
What NN?

Still 43 seconds is a long time. Even a 1993 engines finds the move instantly. Most NN take a long time.
Sorry, should have included that: 384x30-t40-swa-1808000.pb
Top
Hai
Posts: 613
Joined: Sun Aug 04, 2013 1:19 pm

Re: Test position: Neural net blindness.

Post by Hai »

mwyoung wrote: ↑Sat Feb 08, 2020 12:06 pm [d]r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 33

How long does it take for your NN engine to find Rxe6?



New game Line
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Mephisto Gideon Professional (1993) for Winboard and UCI:

33.Rxe6 fxe6 34.Kb2 Qd4 35.Qxd4 Nd5 36.Qd3 Nxb4 37.Qe4 a5
White is clearly winning: +- (10.20) Depth: 12 00:00:23 61592kN
(, 08.02.2020)
LC0 Sergio 40x512 found Rxe6 with 0 seconds.

1.Rxe6 Qxe6 2.Bh3 Qc4 3.Qxf6 a5 4.Bd7 axb4 5.Qb6 bxa3 6.Bc6
-/+ (-1.53) depth: 6 00:00:01 446N

40x512 also found the mate very fast.

How long do we need to wait for the next trained 40x512 net?
Top
peter
Posts: 3252
Joined: Sat Feb 16, 2008 7:38 am
Full name: Peter Martan

Re: Test position: Neural net blindness.

Post by peter »

mwyoung wrote: ↑Sat Feb 08, 2020 12:06 pm [d]r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 33
How long does it take for your NN engine to find Rxe6?
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Fat Fritz v411: (chessbase cloud, 2x2080RTX)

Code: Select all

33.Kb2 De5 34.Dxe5 Txe5 35.Txf6 a5 36.Txf7 axb4 37.Tf4 Td8 38.Ld3 Ted5 39.Txb4+ Ka5 40.h4 Te8 41.Le4 Tc5 42.Td4 Tec8 
  =  (0.04)   Tiefe: 11/22   00:00:01  10kN
...
33.Kb2 De5 34.Dxe5 Txe5 35.Txf6 Te3 36.c3 g5 37.Td6 Txf3 38.Le2 Tf2 39.Td2 Txe2 40.Txe2 Tc8 41.Te7 Td8 42.Kc2 Td6 43.Txf7 Th6 44.Tf2 Kxa3 45.Td2 Tf6 46.g4 h6 47.h3 Ka2 48.Te2 
  =  (0.00)   Tiefe: 19/47   00:00:15  812kN, tb=35
33.Txe6 fxe6 34.Kb2 De5 35.Dxe5 Sd7 36.Dc7 Sb6 37.Dxb6 a5 38.Dg1 axb4 39.Dc5 
  +-  (6.20)   Tiefe: 19/47   00:00:16  942kN, tb=35
...
33.Txe6 fxe6 34.Kb2 De5 35.Dxe5 Sd7 36.Dc3 Sc5 37.bxc5 Td8 38.Db4# 
  +-  (7.16)   Tiefe: 19/47   00:00:23  1758kN, tb=35
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Lc0 v0.23.2 RTX 2080Ti: net 49921

Code: Select all

33.Kb2 De5 34.Dxe5 Txe5 35.Txf6 a5 36.Txf7 axb4 37.Tf4 Ka5 38.axb4+ Kb6 39.Ld3 Td8 40.Tf6+ Kc7 41.Kc3 Td7 
  =  (0.23)   Tiefe: 7/18   00:00:01  2kN
...
33.Kb2 De5 34.Dxe5 Txe5 35.Txf6 a5 36.Txf7 axb4 37.Tf4 Td8 38.Ld3 Ted5 39.Txb4+ Ka5 40.h4 Tf8 41.f4 Te8 42.g4 Te3 43.f5 gxf5 44.gxf5 Tf3 45.Le4 Tdxf5 46.Lxf3 Txf3 47.Td4 
  =  (0.16)   Tiefe: 18/35   00:00:23  863kN
33.Txe6 fxe6 34.Kb2 De5 35.Dxe5 Sd5 36.Dxe6 Sb6 37.Db3# 
  +-  (7.79)   Tiefe: 18/35   00:00:25  942kN
...
33.Txe6 fxe6 34.Kb2 De5 35.Dxe5 Sd7 36.Dxe6 Sb6 37.Db3# 
  +-  (10.06)   Tiefe: 18/35   00:00:30  1217kN
Sorrry for german abbrevations.
Peter.
Top
mwyoung
Posts: 2727
Joined: Wed May 12, 2010 10:00 pm

Re: Test position: Neural net blindness.

Post by mwyoung »

zullil wrote: ↑Sat Feb 08, 2020 12:21 pm Not long:

info depth 12 seldepth 43 time 43748 nodes 721516 score cp 808 wdl 982 7 11 hashfull 2 nps 16509 tbhits 11 multipv 1 pv b6e6 d5e6 b1b2 a8c8 c3b3 e6b3 c2b3

It took mine 19 ply to find the move.


New game Line
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Lc0 v0.23.2+git.c8d9095 Sergio:

33.Rxe6 fxe6 34.Kb2 Nd7 35.Qb3+ Qxb3+ 36.cxb3#
White is clearly winning: +- (8.41) Depth: 19/43 00:00:44 713kN, tb=5
(, 08.02.2020)
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
Top
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: Test position: Neural net blindness.

Post by zullil »

mwyoung wrote: ↑Sat Feb 08, 2020 12:50 pm
zullil wrote: ↑Sat Feb 08, 2020 12:21 pm Not long:

info depth 12 seldepth 43 time 43748 nodes 721516 score cp 808 wdl 982 7 11 hashfull 2 nps 16509 tbhits 11 multipv 1 pv b6e6 d5e6 b1b2 a8c8 c3b3 e6b3 c2b3

It took mine 19 ply to find the move.


New game Line
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Lc0 v0.23.2+git.c8d9095 Sergio:

33.Rxe6 fxe6 34.Kb2 Nd7 35.Qb3+ Qxb3+ 36.cxb3#
White is clearly winning: +- (8.41) Depth: 19/43 00:00:44 713kN, tb=5
(, 08.02.2020)
More or less the same number of nodes.
Top
mwyoung
Posts: 2727
Joined: Wed May 12, 2010 10:00 pm

Re: Test position: Neural net blindness.

Post by mwyoung »

Hai wrote: ↑Sat Feb 08, 2020 12:44 pm
mwyoung wrote: ↑Sat Feb 08, 2020 12:06 pm [d]r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 33

How long does it take for your NN engine to find Rxe6?



New game Line
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Mephisto Gideon Professional (1993) for Winboard and UCI:

33.Rxe6 fxe6 34.Kb2 Qd4 35.Qxd4 Nd5 36.Qd3 Nxb4 37.Qe4 a5
White is clearly winning: +- (10.20) Depth: 12 00:00:23 61592kN
(, 08.02.2020)
LC0 Sergio 40x512 found Rxe6 with 0 seconds.

1.Rxe6 Qxe6 2.Bh3 Qc4 3.Qxf6 a5 4.Bd7 axb4 5.Qb6 bxa3 6.Bc6
-/+ (-1.53) depth: 6 00:00:01 446N

40x512 also found the mate very fast.

How long do we need to wait for the next trained 40x512 net?
You should retest. I already tested both 40x512 nets. And they find it quick for 1 second, and change their minds. And it took a long time again to see Rxe6.
"The worst thing that can happen to a forum is a running wild attacking moderator(HGM) who is not corrected by the community." - Ed Schröder
But my words like silent raindrops fell. And echoed in the wells of silence.
Top
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: Test position: Neural net blindness.

Post by zullil »

mwyoung wrote: ↑Sat Feb 08, 2020 12:55 pm
Hai wrote: ↑Sat Feb 08, 2020 12:44 pm
mwyoung wrote: ↑Sat Feb 08, 2020 12:06 pm [d]r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 33

How long does it take for your NN engine to find Rxe6?



New game Line
r7/5p1p/pR2rnp1/1p1q4/kP6/P1Q2PP1/2P4P/1K3B2 w - - 0 1

Analysis by Mephisto Gideon Professional (1993) for Winboard and UCI:

33.Rxe6 fxe6 34.Kb2 Qd4 35.Qxd4 Nd5 36.Qd3 Nxb4 37.Qe4 a5
White is clearly winning: +- (10.20) Depth: 12 00:00:23 61592kN
(, 08.02.2020)
LC0 Sergio 40x512 found Rxe6 with 0 seconds.

1.Rxe6 Qxe6 2.Bh3 Qc4 3.Qxf6 a5 4.Bd7 axb4 5.Qb6 bxa3 6.Bc6
-/+ (-1.53) depth: 6 00:00:01 446N

40x512 also found the mate very fast.

How long do we need to wait for the next trained 40x512 net?
You should retest. I already tested both 40x512 nets. And they find it quick for 1 second, and change their minds. And it took a long time again to see Rxe6.
Just look at the eval he has.
Top

Return to “Computer Chess Club: General Topics”