CCC Lc0 at position 15

[Jouni]

https://www.chess.com/computer-chess-championship . Yes early days, but still interesting .

[Dann Corbit]

Where do you turn off the sound?
It is easy in TCEC and I do not see any way to do it in CCC.
That "Clack-clack-clack" is annoying to me.

[Chris Formula]

click the speaker icon beside h1

[Dann Corbit]

Thanks, it is watchable now.

I am no longer a fan of unbalanced openings like those used at TCEC and the CCC contests.
In the olden days, we would see real and useful opening novelties that could be easily achieved in game play.
Now, we see which engine plays best from a huge advantage that is not going to happen in real games.

At first, I was a fan because of all the wins and losses.
But now, I wish they would use openings that are achievable in real games in practice.
Maybe they could make a rule that in the final round, if there is no victor after X games, they go to unbalanced.
But if other people like this format, then I submit to the will of the masses.

[Jouni]

Lc0 has improved to position 10-11. Now 35% of games is played. I predict it will NOT be in top 3 positions.

[peter]

After about 1000 of about 3000 planned games

Code: Select all

# PLAYER    : RATING POINTS PLAYED (%)
1 Stockfish : 3713.2 39.0   54     72
2 Torch     : 3694.1 40.5   56     72
3 Obsidian  : 3662.7 35.0   52     67
4 Lc0       : 3623.6 33.5   54     62
...
30Smallbrain: 3407.0 20.5   54     38

[Jouni]

Lc0 ended to position 5. SF, Torch, Berserk and Obsidian were better.

[peter]

Lc0 ended to position 5. SF, Torch, Berserk and Obsidian were better.

The games can be downloaded from archive but the ones shown online are others (a Lc0- Torch bonus- match is shown by title Rating Brawl), with EloStat result looks like that:

Code: Select all

    Program                            Score     %    Av.Op.  Elo    +   -    Draws

  1 Stockfish                      : 109.5/152  72.0   3496   3660   40  37   49.3 %
  2 Torch                          : 109.0/152  71.7   3496   3658   40  38   48.7 %
  3 Obsidian                       : 104.5/152  68.8   3497   3634   40  39   48.0 %
  4 Lc0                            : 102.5/152  67.4   3497   3623   40  39   48.0 %
  5 Berserk                        : 102.5/152  67.4   3497   3623   39  37   52.0 %
  6 Dragon                         :  99.5/152  65.5   3497   3608   40  39   49.3 %
  ...
  ...
  

And with Gamepairs Rescoring Tool from SPCC:

Code: Select all

   # PLAYER         :    Celo  Error   Pairs    W    D    L   (%)  CFS(%)
   1 Stockfish      :    3500   ----      76   67    8    1  93.4      67
   2 Torch          :    3463    168      76   64   12    0  92.1      96
   3 Obsidian       :    3334    161      76   58   15    3  86.2      81
   4 Lc0            :    3277    160      76   54   18    4  82.9      69
   5 Dragon         :    3246    161      76   54   15    7  80.9      50
   6 Berserk        :    3246    159      76   52   19    5  80.9      80
   ...
   ...
   

So at least one thing's clear to be seen: too many engines out of a too inhomogenous pool with much too few games to get out of error bars for a really meaningful (transmissive) rating, even ranking of the best ones isn't quite clear. And then there still is the "weaker" hardware of GPU- side compared to CPU- one relative to TCEC, which matters the more, the bigger the number of CPU- based engines grows against the one and only GPU- based engine. E.g. the wrongly given 1/1 Lc0-Torch- Bonus, the result of which is shown under title Rating Brawl, ended up 52-48% for Lc0 against Torch in head to head match (1 GPU- engine against 1 CPU- engine only) not long ago, regards

[pohl4711]

I am no longer a fan of unbalanced openings like those used at TCEC and the CCC contests.
In the olden days, we would see real and useful opening novelties that could be easily achieved in game play.
Now, we see which engine plays best from a huge advantage that is not going to happen in real games.

At first, I was a fan because of all the wins and losses.
But now, I wish they would use openings that are achievable in real games in practice.
Maybe they could make a rule that in the final round, if there is no victor after X games, they go to unbalanced.
But if other people like this format, then I submit to the will of the masses.

IMHO, here are a lot of misunderstandings:
First: UHO openings (by me or just my UHO-concept (TCEC) (around 50% winning chance for one color)) contain only moves played by humans. My UHO openings are filtered out of the Megabase for example and both players had to have at least 2200 Elo. So, all these UHO lines happened in "real games" (real human games by strong players).

Second: The question, to use UHO or not is not a question of what people like or dislike. It is the question what works and what does not work in these days of superstrong engines. 90%+ draw-ratio is not only boring to watch, but destroys all statistics and measurements.
So, using UHO is the question of "to be or not to be" for modern computerchess. Not, if people like or dislike it.

[Viz]

90% draw ratio is a big understatement, it would be 99% if we go for balaned lines.