Sedat Canbaz Perfect 2019 book. Is it really the issue for Stockfish NNUE?

[mwyoung]

Since some feel that Sedat Canbaz Perfect 2019 book is much too "drawish". And 6 moves is too much theory. And this is not fair for Stockfish NNUE, and skewing the results in my testing.

I generated a opening book to 4 plies. Looking at all positions 160,000+. And had the computer evaluate all positions to a threshold of -33cp at depth 16 plies. For the opening position to be deemed fair.

Let's find out if this theory is true...

Live Stream:

Current standings.

Code: Select all

Result:
------------------------------------------------------------------------------
  #  name              games    wins   draws  losses   score    los%  elo+/-
  1. Lc0 v0.26.2           9       1       8       0     5.0    84.1    38.8
  2. Stockfish 040920      9       0       8       1     4.0    15.9   -38.8

Cross table:
------------------------------------------------------------------------------
  #  name                 score   games         1         2
  1. Lc0 v0.26.2            5.0       9         x ========1
  2. Stockfish 040920       4.0       9 ========0         x

Tech:
------------------------------------------------------------------------------

Tech (average nodes, depths, time/m per move, others per game), counted for computing moves only, ignored moves with zero nodes:
  #  name                nodes/m         NPS  depth/m   time/m    moves     time
  1. Lc0 v0.26.2             89K       20422      9.9      4.3     69.1    299.8
  2. Stockfish 040920    113933K    26845594     39.2      4.2     69.3    294.3
     all ---              55763K    13307606     24.6      4.3     69.2    297.0

[mwyoung]

Update 1

Code: Select all

Result:
------------------------------------------------------------------------------
  #  name              games    wins   draws  losses   score    los%  elo+/-
  1. Lc0 v0.26.2          30       5      20       5    15.0    50.0    -0.0
  2. Stockfish 040920     30       5      20       5    15.0    50.0    -0.0

Cross table:
------------------------------------------------------------------------------
  #  name                 score   games                              1                              2
  1. Lc0 v0.26.2           15.0      30                              x ========1010========10==1010==
  2. Stockfish 040920      15.0      30 ========0101========01==0101==                              x

Tech:
------------------------------------------------------------------------------

Tech (average nodes, depths, time/m per move, others per game), counted for computing moves only, ignored moves with zero nodes:
  #  name                nodes/m         NPS  depth/m   time/m    moves     time
  1. Lc0 v0.26.2             79K       16846     10.1      4.7     57.0    266.7
  2. Stockfish 040920    122954K    26986860     36.8      4.6     57.0    259.7
     all ---              60092K    13322727     23.5      4.6     57.0    263.2