Something new ?

[Rebel]

Something new ?

Hello computer chess friends.

I want to do something new.

I've been a bit annoyed by the one-sidedness of the rating lists for years. It's just about bald elo only. An opening book (as with humans) is an essential part of chess, the same counts for learning, a human does it, a computer can do it but learn files are not allowed at the rating lists. Every (grand)master knows the rating of his opponent and adjusts his strategy, must also be possible, or made possible, for the computer competition. All in all, in the age of the AI, it is about time that computers really start behaving like a human being. Although I don't participate very often, Joost's monthly tournament is a real refreshment.

I want to start a rating list where it is allowed to freely use opening books, learning features, personalities, as long as everything is legal. New creativity, I look forward to it.

I have made preparations, I created an opening set of 1500 positions with moves between 3 and 8, meaning that each engine will at least play 3000 games before it enters the rating list. Games will run on my AMD Ryzen 9 7950X3D 16-Core / 32 threads PC using 16 cores.

Games played will be offered for download, the 1500 opening set will be made public.

Private engines should be sent to me with instructions if needed. Of course your private engines will never leave my hard disk and I will never use any information of your engines. My intent is a new rating list, I will participate with only one version in the beginning at the start, then withdraw to avoid speculations of unfair play.

An example what can be accomplished, I created a learning file and played 2 matches of 1000 games between Stockfish 17.1 and PlentyChess-7.0, the first match using the normal play, the second match giving PlentyChess-7.0 the advantage of the learning file I created.

Code: Select all

No. Name             Win Draw Loss Unf.  Score Games       %
------------------------------------------------------------
  1 Stockfish-17.1  +361 =467 -172   *0  594.5  1000   59.5%
  2 PlentyChess-7.0 +172 =467 -361   *0  405.5  1000   40.5%

Code: Select all

No. Name             Win Draw Loss Unf.  Score Games       %
------------------------------------------------------------
  1 PlentyChess-7.0 +292 =490 -218   *0  537.0  1000   53.7%
  2 Stockfish-17.1  +218 =490 -292   *0  463.0  1000   46.3%

A difference of 91 elo.

Let's shake that tree!

New creativity.

[Ajedrecista]

Hello Ed:

That learning file is very worthy! I calculate circa +92.22 Elo taking decimals into account. Rounding to the nearest 0.01 Elo, I calculated from SF POV and 95% confidence, using the trinomial distribution model: +66.46 ± 15.77 Elo and -25.76 ± 15.40 Elo (in fact, there are assymetric intervals -15.63, +15.90; -15.45, +15.35). In other words: [+50.83, +82.37] Elo and [-41.21, -10.41] Elo, respectively.

If we combine the two tests, which should be a statistical nonsense due to the different conditions: +20.00 ± 11.00 Elo (assymetric intervals of -10.98, +11.02). In other words: [+9.02, +31.02] Elo.

It would be interesting to revert the experiment: how about PlentyChess without learning file; and SF with and without learning file?

Regards from Spain.

Ajedrecista.

[Graham Banks]

Hi Ed,

the only issue with learning is that your Plentychess might learn differently to that of other users, due to the different openings they are likely to use.

[Rebel]

Hi Ed,

the only issue with learning is that your Plentychess might learn differently to that of other users, due to the different openings they are likely to use.

The learning file I created works for every engine.

But that's not the topic.

The only reason I am taken this initiative is that the well known rating lists (such as yours) refuse engines to play like humans.

[Graham Banks]

Hi Ed,

the only issue with learning is that your Plentychess might learn differently to that of other users, due to the different openings they are likely to use.

The learning file I created works for every engine.

But that's not the topic.

The only reason I am taken this initiative is that the well known rating lists (such as yours) refuse engines to play like humans.

Well it's always nice to have additional rating lists.

[Jouni]

So learning file is opening book? PlentyChess has no learning.

[Modern Times]

The only reason I am taken this initiative is that the well known rating lists (such as CCRL) refuse engines to play like humans.

Well, apart from the fact that the rules of the game are the same, I see human chess and computer chess as very different things.

And there are things like DFRC, Chess324 that are an invention purely for Computer chess.

[chrisw]

Hi Ed,

the only issue with learning is that your Plentychess might learn differently to that of other users, due to the different openings they are likely to use.

The learning file I created works for every engine.

But that's not the topic.

The only reason I am taken this initiative is that the well known rating lists (such as yours) refuse engines to play like humans.

Well it's always nice to have additional rating lists.

It would be good to have a community rating list, or just alternative rating lists, because the one that managed to get itself branded is run by the (Moderation: removed content that was not in line with the charter, 6. Are not bullying, or vicarious bullying) person whose attitude is because “I say so” (Moderation: removed content that was not in line with the charter, 6. Are not bullying, or vicarious bullying).

[mar]

Hi Ed,

the only issue with learning is that your Plentychess might learn differently to that of other users, due to the different openings they are likely to use.

The learning file I created works for every engine.

But that's not the topic.

The only reason I am taken this initiative is that the well known rating lists (such as yours) refuse engines to play like humans.

Well it's always nice to have additional rating lists.

It would be good to have a community rating list, or just alternative rating lists, because the one that managed to get itself branded is run by the (Moderation: removed content that was not in line with the charter, 6. Are not bullying, or vicarious bullying) person whose attitude is because “I say so” (Moderation: removed content that was not in line with the charter, 6. Are not bullying, or vicarious bullying).

wow, what an incredibly arrogant post

[lucario6607]

Something new ?

Hello computer chess friends.

I want to do something new.

I've been a bit annoyed by the one-sidedness of the rating lists for years. It's just about bald elo only. An opening book (as with humans) is an essential part of chess, the same counts for learning, a human does it, a computer can do it but learn files are not allowed at the rating lists. Every (grand)master knows the rating of his opponent and adjusts his strategy, must also be possible, or made possible, for the computer competition. All in all, in the age of the AI, it is about time that computers really start behaving like a human being. Although I don't participate very often, Joost's monthly tournament is a real refreshment.

I want to start a rating list where it is allowed to freely use opening books, learning features, personalities, as long as everything is legal. New creativity, I look forward to it.

I have made preparations, I created an opening set of 1500 positions with moves between 3 and 8, meaning that each engine will at least play 3000 games before it enters the rating list. Games will run on my AMD Ryzen 9 7950X3D 16-Core / 32 threads PC using 16 cores.

Games played will be offered for download, the 1500 opening set will be made public.

Private engines should be sent to me with instructions if needed. Of course your private engines will never leave my hard disk and I will never use any information of your engines. My intent is a new rating list, I will participate with only one version in the beginning at the start, then withdraw to avoid speculations of unfair play.

An example what can be accomplished, I created a learning file and played 2 matches of 1000 games between Stockfish 17.1 and PlentyChess-7.0, the first match using the normal play, the second match giving PlentyChess-7.0 the advantage of the learning file I created.

Code: Select all

No. Name             Win Draw Loss Unf.  Score Games       %
------------------------------------------------------------
  1 Stockfish-17.1  +361 =467 -172   *0  594.5  1000   59.5%
  2 PlentyChess-7.0 +172 =467 -361   *0  405.5  1000   40.5%

Code: Select all

No. Name             Win Draw Loss Unf.  Score Games       %
------------------------------------------------------------
  1 PlentyChess-7.0 +292 =490 -218   *0  537.0  1000   53.7%
  2 Stockfish-17.1  +218 =490 -292   *0  463.0  1000   46.3%

A difference of 91 elo.

Let's shake that tree!

New creativity.

how do you use a learning file. From my experience shashchess learning doesn't do much?