Winter is Coming (Amoeba, Beef, Combusk, Demoli, Fab, Halogen, Koivi, Lc0, Marvin, Minic, Nemori, Orion, Pedone, Winter)

[jorose]

Thank you for the tournament!

Indeed luck for sure plays a big part in 58 rounds. From the rating lists we know the two Winters are closer to about 100 Elo separation whereas here the performance differs by about 170.

I quite enjoyed following the classic old-school large double round robin format. In the past a lot of the supertournaments had this format. Nowadays you just have Wijk/Tata Steel doing something similar, and even there, 14 players is somewhat constrained.

[Angle]

Indeed luck for sure plays a big part in 58 rounds. From the rating lists we know the two Winters are closer to about 100 Elo separation whereas here the performance differs by about 170.

Probably, the reason is in a quite fast TC and the fact that search/depth/precision was greatly improved in ver. 0.9. In some games that I have observed Winter 0.8 launched promissing (though risky) attacks according to its agressive playing style but in subsequent complicated game with widely open positions its search depth dropped dramatically (aggravated by the lack of time). So, it was out-performeded by faster and qualified opponents. On the other hand, Winter 0.9 in performing similar attacks demonstrated much better search rate and precision (even when time was limited) which allowed it to outplay the opponents.

I quite enjoyed following the classic old-school large double round robin format. In the past a lot of the supertournaments had this format. Nowadays you just have Wijk/Tata Steel doing something similar, and even there, 14 players is somewhat constrained.

Initially I planned *only* 24 participants, so the decision to expand the tourney was quite spontaneous
Although I also like large double round robin tourneys , the amount of 870 games is quite fatiguing both for me and my laptop

[mar]

Thanks for the tournament!

not great, not terrible for Cheng (against stronger engines).
only 58 games but perhaps dev is somewhat stronger indeed, but seems to risk a lot more. needs more work for sure.

Initially I planned *only* 24 participants, so the decision to expand the tourney was quite spontaneous
Although I also like large double round robin tourneys , the amount of 870 games is quite fatiguing both for me and my laptop

Well, maybe it would be better to reduce the number of engines of the same breed to say 2 versions?
Too much Beef, Demolito and Marvin for my taste

[Frank Quisinsky]

Hi Martin,

style is more important as Elo.
You can be sure, you developed a strong and very nice chess engine.

Hope you have fun on further developments!

In my opinion, all is right with Cheng4.



Thanks for Cheng4 and the new dev version!!

Best
Frank

[Angle]

Well, maybe it would be better to reduce the number of engines of the same breed to say 2 versions?
Too much Beef, Demolito and Marvin for my taste

In general I agree, but ... For Demolito, I had a special reason to unclude so many versions because three of them were ''undertested'' in my personal rating list. So I needed some extra games by them to get more reliable results (ratings). Initially, I included two Marvin versions: 3.6.0 and 4.0.0. However, ver..4.0.0 played with some bugs, so during the tourney Martin released a bugfix version 4.0.1 which I decided to include too. Finally. Beef initially was presented only by version 3.0.6, but Frank suggested to include 0.2.2 (to compare the results). I also included ver. 0.2 since all three Beef versions were nearly equal in my previous tests (the other reason was to get even number of participants).

[mar]

Thanks Frank,

I believe that style comes from search and eval inaccuracies, after all, eval features are tuned automatically.

I always find it amusing when someone claims that his engine plays a very human style(tm), when it's really auto-tuned
based on a ton of self-play games. placebo is strong!

Best

Martin

[Frank Quisinsky]

Hi Martin,

oh the reason for it is very easy ...

Humans have no chance vs. strongest computer chess engines and the reason is the endgame.
In the earlier middlegame with a lot of knowledge to opening systems strongest humans can hold the game.

With lesser pieces on board no chance for humans.

If we have engines do the following things:

- very strong with great king safty after opening books moves ...
- hold the style in the earlier middlegame
- lost strenght in the late middlegame and endgame

We have the human style generally.
Quit clear that humans plays different styles of chess.
But generally, not important which style of chess plays humans ... because which each of the possible styles humans generate no chance vs. strongest engines in late middlegame, transposition into endgame and endgame.

Wasp for an example play the "human style" generally.
With very aggressive pawns simulate Wasp in additional an aggressive style with many pieces on board after the openings!

It's much more easy for me to find out the style from an engine as for one of the 16. World Champions!
Statistic can speak with us, if we like to have a look on it.

Examples: The bad king safty from Rybka with many pieces on board. The very speculative style from Fizbo ... can produce a lot of fast won games but produce a lot of fast lost games too. That is the secret! I think the reason why persons like to have many chess engines on the hard disk. If all the engines do the same ... we all have a very boring hobby. So each chess program produce an own style after the work the programmer do. With or without autotunings!

The problem is:
The more stronger an engine is ... the more complicated is to find out the strengths and weaknesses. And without to find out strengths and weaknesses no possibility to build a playing style. That's really a problem for me ... if I looking in Stockfish ... I see nothing. All is strong what Stockfish like to play. I can't build a special style.

Have a look in Ktulu ... Rahman with his rook endgames. Without endgame bases Ktulu can play 70% of rook / pawn endgames in perfection with special knowledge. The Tornado programmer and the strong endgame ... around the same style Ktulu have. No king safty for many pieces on board and a lot of fast lost games.

All this I like so much!
If the programmer are able to give an engine an own face!

And I believe the own face is possible with strength and weaknesses we can see.
And for this reason ... no chess program must play perfect chess. If we have a handfull of such programs ... it's enough for me. Speaking for myself.

Best
Frank

[Frank Quisinsky]

Another good example is Winter:
Why Winter 0.8 lost so many games in the late middlegame.
No other engine with such a high playing strength do that.

All the fast lost games from Rybka comes more from the first middlegame phase.
Winter have a problem in the second middlegame phase.

That's not bad for me ... that's great!!
A new variant!!
And after a good first playing phase and the late middlegmaes with problems ... Winter can play a fantastic endgame!!

Can be see in statistics!
Have a look on FCP Tourney-2020 stats.
Will be better if Klaus working on it if FCP Tourney-2021 is still running.
For the moment we are K.O.



Now, if the programmer of Winter fix such problems and Winter goes to 3.400 Elo like Stockfish what will be the final result?
The next engine we can see nothing, perfect in all the playing phases.

If computer chess goes this way all will be more and more boring!
I think the secret today is ... the programmers should try to give engines a clear own face.
That's much more interesting as Elo or autotunings again and again!

Back to the roots with pure chess knowledge can be very interesting! All the different playing styles in chess computer times, later in DOS / Winboard times. And the people like that!!

Rahman do that in the past with his main interest: How I can try that Ktulu play the endgame with rook and pawns not to 72% perfect ... how 75% are possible? The big topic from the programmer from Iran ... for an example. Whatever Ktulu like to play in the earlier middlegame ... not important for him.

Best
Frank

Winter or Cheng, Ktulu or Wasp ...
With all of the engines I have fun if I can see that the engines like to play another style.
Komodo or Stockfish ... I can looking on test-results and Elo only and understand nothing!
So what is the more interesting chess programm ... a question for you as programmer of Cheng!?

If your answer is Stockfish ...
The next question ... Do you like to see 50 programs with the same strength of Stockfish?

[Frank Quisinsky]

Special programmings in set weaknesses for very strong engines can be interesting!

If very strong programs like Stockfish will give me the chance to set weaknesses via UCI settings ...

We can simulate each chess player if we find out the weaknesses from a special chess player!

The secret today is to make the strongest engines weaker, not stronger with special features.
So more chess player like to install engines for self playing and trainings and computer chess will go in the next aera.
But perhaps not the right time to do that, most are thinking ... how I can make my program stronger.

Will be a main topic for the next years.
To set the strength from 100% to x% is to easy.
I think our chess programmers can do much more.



Best
Frank

[mar]

Winter or Cheng, Ktulu or Wasp ...
With all of the engines I have fun if I can see that the engines like to play another style.
Komodo or Stockfish ... I can looking on test-results and Elo only and understand nothing!
So what is the more interesting chess programm ... a question for you as programmer of Cheng!?

well I can give you a list of engines I like, in no particular order:
Dirty, iCE, Gaviota, Baron, Texel, Critter, Rybka (people will hate me for it but 3,4 is a masterpiece), Isa, AnMon, Hermann and I could continue on and on. there are many interesting engines out there and I'm sure I forgot many I wanted to mention

but nobody gives a damn because people demand new and strong engines - that's all they really care about and I can understand that.
engines that are 3+ years old (no new version since then) - nobody cares about them anymore

at a certain point chessprogramming boils down to 99.9% testing and 0.1% programming - not much fun, at least not for me