Openings used in CCRL

Thx, I need to check your suggestions, Frank. Currently I´m rather using Ed´s engines from his site (Rebels, ProDeo, Benjamin -- really great stuff, very much appreciated!) or old Shredder (7, 9, 10), early Komodos or even Lc0 with Bad/Evil Gyal nets with white -- against more recent engines like Fat Fritz, Lc0-J92-270, Hiarcs 14 or CF with black for the start. Openings are at the moment Philidor and the Dutch Leningrad system.

Nez, Bozo sounds not too bad for blitz indeed I suppose but then you need to know your lines in Nimzowitch and Chigorin, as you say. My approach is rather to play 1.- d6 and then the above ones or even Pirc, KID or Old Indian. In particular the latter is similar to the Philidor, so over time you get used to the typical set-ups and ideas. Nevertheless tiny and subtle differences can have significant effects, e.g. the order white is playing h3 or a4 or just one of them. It seems that most engines don´t like the black side and don´t play it really well (need to check out SlowChess though).

Additinally, if you let play the strongest engines stuff like that, then you see many drawn games with an occasional black loss. Well, to be honest: that´s not what I´m going for... And it´s not my experience against human players either!

So, what I´m doing is: choose complex openings like the above (can be unbalanced, -0.5 or so by SF standards) and give the white side to 2300-2800 engines. Now, which engine with black, except the usual suspects, can reach a satisfactory position after, say, 30-40 moves? I don´t want to see which engine can win its endgames best, that´s a different question. What I´m interested in is, what are the most promising ways to stirr up the game and be still ok. I find this approach very entertaining but I understand that it might not be everybody´s cup of tea.  

Cheers
Martin

Hi Martin,

you don't need the unbalanced openings.
That's my personal opinion after so many years I am working on the FEOBOS project!

Interesting is more to find out which of the balanced openings produced a higher draw quote. For give answeres a lot of games with the FEOBOS databases are necessary. If I have enough material collected I think I can start a second FEOBOS project.

Means all what is interesting is inside FEOBOS.
I check that often with newer GM theory, opening books (literature) Encyclopaedias and all material I can find. But 3 moves only after ECO codes formed ... new material will be not found. I analyzed that again and again before FEOBOS started and during FEOBOS is still running.

The high draw quote from best engines have in clearly most of cases other reasons!
The style is often the same and the strength for endgames is the same.
The higher draw quote comes not from the openings we can manipulated.
Have other reasons!

The draw quote comes from openings is select out to more as 50% with the Contempt factor we set in FEOBOS, speaking from the fast 3-fold draws.

Let me look in FEOBOS what we do with SF:
http://www.amateurschach.de/common/feobos/excel/settings.png

-0.30 for Stockfish!

The problem is here that all the different engines comes with different evaluation. iCE or Texcel others give faster a 0.8 as others for one example. With Excel easy to find out if we put a template over 41.614 positions.

:-)

A lot of Philidor lines are not easy to understand for humans and for myself. Complicate opening for myself. I try to avoid Philidor. Dutch Leningrad system I like. I am playing with black after 1. e4 in 90% of cases 1. ... d5 and after 1. d4 in 90% of all cases 1. ... f5.

:-)

Best
Frank

Openings and mid-games is most interesting in computer chess for myself.

Hi Frank,

not sure what the FEOBOS project is doing exactly but looks like quite lot of work. Of course there are probably various reasons for the higher draw rates: not only the really strong engines in the mean time but also the choice of openings. And what we humans find interesting might be drawish for the engines...

Coming back to the original posting of this thread: One could argue that “only” 1.6% of the games (~19k) outside of a [-0.5,+0.5] eval window out of the opening, nevertheless leading to about 60% decided games in total (>700k), is not that bad (from https://www.computerchess.org.uk/ccrl/4040/games.html ). Many equal evals from the opening lead to decided games anyway.

What I´d find interesting though is the question if the more recent vintages show a different pattern. If we take e.g. the 257.040 games from Dec´18-Nov´20, do we see a different draw rate compared to the two years before that? And is the percentage of “suspect openings” in this sense the same? I would choose two year time windows because not much earlier Leela “by September 2018 (…) had become competitive with the strongest engines in the world”, according to a claim from the Wiki, pushing the playing level in general much higher until today. Assuming that the opening choices are not that much different on average, however, any changes in draw rates could be attributed to stronger play in general. And in the games played in 2006-2010 the top engines were probably up to ~500(?) Elo lower than today.

Yes, around 500 Elo lower than today (strongest engines today like Stockfish, Komodo or LcO).
Yes, roughly we could say that the openings are the same. In detail we have better possibilites to find the opening blunders for engine books or such databases FEOBOS is.

Yes, the draw quote is higher as in the past.

In my opinion the high strength today comes from the transposition into endgames. If many programs have the same strength and weaknesses is this the same as to play Ruffian vs. Ruffian for 18 years or Fritz 8 vs. Fritz 8. We produced a higher draw quote.

More Elo generally = higher draw quote ... yes, a bit higher!
Same strength and weaknesses = a clearly higher draw quote.

That's my opinion!
Perhaps forced the higher Elo the playing phase "transposition into endgame" and late mid-game generally with the possibilities the programmers have today. I can't see that the strongest engines give always best moves after the openings or with many pieces on board.

To analyze engines like Stockfish isn't easy because. Not enough different opponents are available to create better statistics.

Again to the openings:
With the strengths from best engines today it's maybe possible to find the openings where the probably for a draw is higher. That's one of the interesting points in computer chess today.

Often I can speculate only, not important how many statistics I read or created.

Example:
Wasp 4.00 have 3000 Elo = fact.
We are to 100% sure (fact), that Wasp lost a lot of Elo in endgames if we looking in statistics.

Situation:
Wasp 4.00 with 3000 Elo was available for 10 years.
Others programs for 10 years max. with 2800 Elo.

No person in the World are able to find that Wasp with 3000 Elo have problems in endgames.

We had such situations with Rybka or Shredder!
Rybka was the number 1, Shredder the number 1 in the past and both programs have a really bad king safty in midgames. Today we know that because we have stronger opponents and enough game material. But in that times Rybka on 1, Shredder one 1, nobody find out that ... why?

Nobady have 3000 Elo.
So, we can work with statistics only!!

Maybe:
If we have more engines with different styles of chess, but around the same Elo the draw quote is lower as today for matches Stockfish vs. Komodo (for an example).

Best
Frank