FCP Tourney-2022 ...

[carldaman]

And much clones seems to be stronger as Stockfish is.
Same problem for the current dev. in the still running tourney.

A good chance for Komodo to be the new number 1 with longer time controls and contempt!!

For some years I test it out two times (SF with and without contempt) vs. much weaker engines.
- 19 Elo in test-1 after 1.500 games
- 17 Elo in test-2 after 1.500 games

Stockfish lost around 18 Elo without contempt in test-runs with more as 30 opponents.
Maybe the same today?!!

Fact ist, that Engines like Stockfish and Komodo lost a lot of Elo if we put more weaker opponents
in a test-run.

Can be see if we compare CEGT / CCRL Elo with FCP Tourney-2021, FCP Tourney-2020 and of course
with FCP Tourney-2022. With the Excel from Klaus we can deleted engines in the field of oppoents.
With that feature it should be more clear.

I'm a strong advocate of including more weaker opponents in test-runs for this reason.
It will also alleviate the draw death that plagues tourneys and matches with only the strongest engines
and also reward those getting more wins with more Elo.

The counter-argument is that including much weaker engines will cause rating distortions, but I ask, wouldn't avoiding these engines cause a rating distortion for a top engine SF, also, by over-inflating its rating?

The fact that SF draws too much against weaker opposition should be reflected in its rating, or else that rating is inaccurate. An engine with smart contempt like Komodo should definitely benefit, as long as its contempt is set at a reasonable level.

You could test that hypothesis in your tournaments, Frank!

[connor_mcmonigle]

And much clones seems to be stronger as Stockfish is.
Same problem for the current dev. in the still running tourney.

A good chance for Komodo to be the new number 1 with longer time controls and contempt!!

For some years I test it out two times (SF with and without contempt) vs. much weaker engines.
- 19 Elo in test-1 after 1.500 games
- 17 Elo in test-2 after 1.500 games

Stockfish lost around 18 Elo without contempt in test-runs with more as 30 opponents.
Maybe the same today?!!

Fact ist, that Engines like Stockfish and Komodo lost a lot of Elo if we put more weaker opponents
in a test-run.

Can be see if we compare CEGT / CCRL Elo with FCP Tourney-2021, FCP Tourney-2020 and of course
with FCP Tourney-2022. With the Excel from Klaus we can deleted engines in the field of oppoents.
With that feature it should be more clear.

I'm a strong advocate of including more weaker opponents in test-runs for this reason.
It will also alleviate the draw death that plagues tourneys and matches with only the strongest engines
and also reward those getting more wins with more Elo.

The counter-argument is that including much weaker engines will cause rating distortions, but I ask, wouldn't avoiding these engines cause a rating distortion for a top engine SF, also, by over-inflating its rating?

The fact that SF draws too much against weaker opposition should be reflected in its rating, or else that rating is inaccurate. An engine with smart contempt like Komodo should definitely benefit, as long as its contempt is set at a reasonable level.

You could test that hypothesis in your tournaments, Frank!

FCP already includes many weaker engines, no?

[Frank Quisinsky]

Hi Conner,

that's right.
No weaker engines, the TOP-41 ... bad word by myself.
Weaker engines from the view of the number 1 or number 2.

Interesting is ...
10 months later to FCP Tourney-2021 ...
If Shredder 13 = ~ 3125 Elo ...

1. FCP Tourney-2021 = 01. Stockfish 110121 NN x64 : 3462 ... 40 moves in 12 minutes on 4.9Ghz ... 2.000 games
2. FCP Tourney-2022 = 02. Stockfish 151121 NN x64 : 3434 ... 40 moves in 20 minutes on 4.4Ghz ... 173 games

Sure, Stockfish is stronger 10 months later.
Most of opponents are clearly stronger too, 2022er tourney have a average ~ +80 to 2021er tourney.

OK, not enough games at the moment.
But to compare both tournaments later can be very interesting!

Best
Frank

BTW:
Seer is very aggressive with many pieces on board.
Will be good to see with stats we will make later ... like that!!

[Frank Quisinsky]

Hi there,

I have to stop the tourney again!

Wasp 5.0 NN will be available today and games from Wasp 4.66 NN dev. will be replayed.
Wasp 5.0 NN release will be around 25 Elo stronger as Wasp 4.66 NN dev. and it could be possible +75 Elo or a little bit more to Wasp 4.50.

Later ...

I will start working at first on the tournament files and at second on the release information on my webpage!



Best
Frank

[carldaman]

And much clones seems to be stronger as Stockfish is.
Same problem for the current dev. in the still running tourney.

A good chance for Komodo to be the new number 1 with longer time controls and contempt!!

For some years I test it out two times (SF with and without contempt) vs. much weaker engines.
- 19 Elo in test-1 after 1.500 games
- 17 Elo in test-2 after 1.500 games

Stockfish lost around 18 Elo without contempt in test-runs with more as 30 opponents.
Maybe the same today?!!

Fact ist, that Engines like Stockfish and Komodo lost a lot of Elo if we put more weaker opponents
in a test-run.

Can be see if we compare CEGT / CCRL Elo with FCP Tourney-2021, FCP Tourney-2020 and of course
with FCP Tourney-2022. With the Excel from Klaus we can deleted engines in the field of oppoents.
With that feature it should be more clear.

I'm a strong advocate of including more weaker opponents in test-runs for this reason.
It will also alleviate the draw death that plagues tourneys and matches with only the strongest engines
and also reward those getting more wins with more Elo.

The counter-argument is that including much weaker engines will cause rating distortions, but I ask, wouldn't avoiding these engines cause a rating distortion for a top engine SF, also, by over-inflating its rating?

The fact that SF draws too much against weaker opposition should be reflected in its rating, or else that rating is inaccurate. An engine with smart contempt like Komodo should definitely benefit, as long as its contempt is set at a reasonable level.

You could test that hypothesis in your tournaments, Frank!

FCP already includes many weaker engines, no?

Yes, I was (indirectly) agreeing with his concept for a tourney.

[Frank Quisinsky]

done ...

- Wasp 5.00 NN is still running for FCP Tourney-2022.
- Wasp 5.00 NN released.

Before I start the work with Klaus Wlotzka for statistics ...

Engine updates possible until November 21st, 2021 (23:45) are possible!

Go Wasp go ...


Hint for John Stanback:
The bench results from the three FCP Tourney-2022 systems can be found in Wasp 5.00 download file and on the detail page to the tournament:

FCP Tourney-2022, detail page:
https://www.amateurschach.de/main/_fcp-tourney-2022.htm

FCP Tourney-2022, engine configuation (if an user need help or a programmer have interest to check that):
https://www.amateurschach.de/main/_conf ... n-2022.htm

And again ...
Good luck to all participating programmers!

Best
Frank

[Frank Quisinsky]

Conner, Carldamen,

for some years (working on FCP Rating List, or working before on SWCR Ratinglist) I had the opinion, after many stats I made for test-runs, Elo different between strongest and weakest engines should be not more as 280 points ... if I added min. 26 opponents.

FCP Tourney-2022 = Between strongest and weakest engine = ~ 425-450 Elo (41 partipating engines).

Not important here ...
Each one against each other, 50 games per match!
Or the time control, other things!

The final results can't be exactly.

But with all the other ideas I had in so many years computer-chess ...
The final Elo result will be never exactly.

The reason that Elo is for myself not very important.
Much more important is to find out a bit about the "playing styles" of engines!

And for this reason I need ...
Each one against each other, 50 games per match!

I am sure with Klaus we can create some strong statistics during the FCP Tourney-2022 is still running.

In one of the statistics we can see the fast draw games.
We can compare Stockfish without contempt with Komodo with contempt (for one example).

So, contempt parameter not needed for testing such things!

Example:
In my opinion the fast draws from Wasp 4.50 in FCP Tourney-2021 are to high.
Wasp 4.50 is playing with Contempt = 0.
But it was the wish from John not to changed the contempt parameter.

Wasp 5.00 is running with default settings too (with Contempt = 0).

Again, later we can see very easy in our statistics which engines should play better with contempt = x.

Best
Frank

[Frank Quisinsky]

Hi there,

the next update:

SlowChess Blitz 2.8 replaced SlowChess Blitz 2.7.

Go Slow go ...



Best
Frank

[carldaman]

Conner, Carldamen,

for some years (working on FCP Rating List, or working before on SWCR Ratinglist) I had the opinion, after many stats I made for test-runs, Elo different between strongest and weakest engines should be not more as 280 points ... if I added min. 26 opponents.

FCP Tourney-2022 = Between strongest and weakest engine = ~ 425-450 Elo (41 partipating engines).

Not important here ...
Each one against each other, 50 games per match!
Or the time control, other things!

The final results can't be exactly.

But with all the other ideas I had in so many years computer-chess ...
The final Elo result will be never exactly.

The reason that Elo is for myself not very important.
Much more important is to find out a bit about the "playing styles" of engines!

And for this reason I need ...
Each one against each other, 50 games per match!

I am sure with Klaus we can create some strong statistics during the FCP Tourney-2022 is still running.

In one of the statistics we can see the fast draw games.
We can compare Stockfish without contempt with Komodo with contempt (for one example).

So, contempt parameter not needed for testing such things!

Example:
In my opinion the fast draws from Wasp 4.50 in FCP Tourney-2021 are to high.
Wasp 4.50 is playing with Contempt = 0.
But it was the wish from John not to changed the contempt parameter.

Wasp 5.00 is running with default settings too (with Contempt = 0).

Again, later we can see very easy in our statistics which engines should play better with contempt = x.

Best
Frank

[It's probably best to test with default contempt, unless you're testing a different setting that interests you.]

As Brendan Norman can probably tell you as well, the true character, or 'inner soul' if you will, of an engine only completely comes out when playing down a few hundred Elo, in terms of opposition. Then you get to find out how
an engine really plays, when it's expected to win. Of course upset wins against stronger opponents can provide a lot of excitement, too, but those are far less common or typical.

So, a wide variety of opponents with different strengths and styles is a really good thing in that regard if you are someone who also cares about the aesthetics of the game.

[Frank Quisinsky]

Hi,

yes I like to read the messages by Brandon and like what Brandon give for inforamtion.
And yes, the playing style is all of time most important for me. For that reasons are engines like AnMon, Zarkov, Phalanx, Patzer in Winboard times most interesting for me. Aristarch, ETChess, Gromit / Anaconda and different others.

Patzer with his aggressive free pawns.
AnMon like to play b4, g4, b5, g5 moves in closed positions with many pieces on board. With the result that interesting mid-game positions will be produced AnMon like it to play.

Phalanx was a great king attacker and Zarkov try to create an attack only ... if the own king is save with strategical ideas.
All the time John's programs have a good king safty.

If we are looking today in stats to the playing strength often we can see that most of strength comes from the passage ... end of mid-game to endgame. A lot of chess programs playing here near the perfection but I missed the aggressive style in the earlier mid-game. So most interest I have to pick-up the engines can give us information to A00-E99 with newer ideas. That's for humans very interesting.

Have a look in the complicated Dutch openings or kings-indian openings. Slow is the Dutch-Master of engines for an example.

It make sense to create an opening book with equal positions for most complicated opening systems and to play with most aggresssive engines "eng-eng" tournaments with longer time-controls. Often I am thinking about it to do that but to many holes in own files. At first I need a new FEOBOS project ... but this is a lot of work. But this was the main reason I ordered the AMD Ryzen 9 systems ... main reason was not the FCP Tourney-2022.

Very special what I like to see but I am sure ...
We all are very special for different reasons around what we like to see in computer-chess.

Best
Frank