Mystery engine at CCC

[Ras]

Unbalanced openings are perfectly fine if each engine gets to play it with both sides against the same opponents. A better engine might convert the advantage into a win or hold the disadvantage to a draw, compared to drawing with advantage and losing with disadvantage. It just shouldn't be so unbalanced that the results are obvious, such as starting with a piece handicap.

[CornfedForever]

For once, I agree with Eduard.

Have you...been to see a doctor about that?

[Eduard]

<snip>The creation of a whole new engine is actually completely unnecessary.</snip>

Then ignore it. Nobody is forcing you to use it. It sounds like Stockfish is all you need, and objectively that’s probably true.

— Steve

I prefer my own GUI (I am using ChessBase). An engine that I cannot install under my GUI will always remain of no interest to me. The bullet strength of an engine may be good for rankings, but I know that other criteria count for professional players. In the last months I got in contact with strong correspondence chess players, also with top players. That's how I know that Stockfish dev is not the No. 1 engine for such players. Many prefer Stockfish clones. Because Stockfish is open source, you can do a lot with Stockfish. There are Stockfish clones that are very strong in tactical positions (Crystal), and there are clones that use a very interesting learning file, eg. Eman. Because of the learning file (also works during analysis) Eman is very popular. Another interesting engine is CorChess. This Stockfish clone is clearly better than Stockfish dev in long analysis.

I implemented Eman's learning code in Stockfish for myself, and also modified the search. One of these engines uses CorChess search paired with Eman's learning file. Eman's learning code is included in the StockfishMZ open source engine.

Some of my correspondence chess friends use my Stockfish clone. Only yesterday a top ICCF player (board 1) told me that he uses my engine as the main engine for analysis. I prefer using CorChess to Stockfish dev, and if I can implement Eman's learning code as well, that's very cool.

It doesn't matter who does it either. I just want to say: There are many ways to modify Stockfish to your liking. There isn't just ONE Stockfish dev. Depending on the position type, I use different Stockfish clones. Correspondence chess players love the learning file and that's what the clones are for!

For professional chess players, not only the pure tester ranking counts. A good engine should also contain useful features, e.g. a learning file. Therefore it is important that the engine can also be installed in the preferred GUI. Everything else, like online analysis on servers, is childish stuff.

[lkaufman]

If you don't use unbalanced opening books, it will be all draws I'm afraid........

With top engines only, more like around 80% draw rate.
For other engines, you obviously get a lower to much lower draw rate.

CCRL 40/15 uses fairly balanced books:

For Dragon 3.2 4 cpu on CCRL Rapid I count about 96% draws against the seven opposing engines over 3500 (all SF or SF derivatives of course). Most likely the few decisive games were from openings that would now be considered somewhat dubious, even if they were thought to be reasonably balanced when the books were made. If you use books that never exit with Black more than say 5 centipawns worse than "par" (the initial eval) I think it would be over 99% draws. There are no other valid pairings over 3500 that are not between SF derivates or other Dragon versions.

As long as an evaluation dips below 0.70 for even one move for the first 10 moves out of book, I'm happy.

I agree with you that an eval of less than 0.70 from the opening is desirable, I just would characterize any evals above 0.50 as "unbalanced". No top human GM would intentionally play a defense that lead to a score worse than 0.50 against another top GM in an important classical game except perhaps for surprise value or due to needing to win with Black due to match score. Only rarely do they choose defenses that are worse than -.40. Top engines would rarely choose defenses worse than about -.30 on their own, even with MP randomness. But it is necessary to include openings that are close to your 0.70 line in order to avoid draws. You are in effect just doing a mild version of the UHO idea when you do this, which I think is fine. It's a good compromise between "correct" chess and the need to avoid 100% draws.

[Graham Banks]

For once, I agree with Eduard.

Have you...been to see a doctor about that?

[Eduard]

Unbalanced openings are perfectly fine if each engine gets to play it with both sides against the same opponents. A better engine might convert the advantage into a win or hold the disadvantage to a draw, compared to drawing with advantage and losing with disadvantage. It just shouldn't be so unbalanced that the results are obvious, such as starting with a piece handicap.

It's strange nonetheless. An engine that only gets 50% with normal openings but has +100 Elo with unbalanced openings, touting it as a 100 Elo better engine is strange. Then there is the question of what it looks like in position tests. If the +100 Elo engine is not better in position tests, then the 100 Elo makes absolutely no sense and suggests a distorted playing strength. For me personally, +2 Elo with normal openings count more. For this I have a test suite for 200 games with Ponder ON plus a position test suite, and then I go onto the server and play a few hundred games live. But to each his own, good luck with Torch!

[Peter Berger]

I agree with you that an eval of less than 0.70 from the opening is desirable, I just would characterize any evals above 0.50 as "unbalanced". No top human GM would intentionally play a defense that lead to a score worse than 0.50 against another top GM in an important classical game except perhaps for surprise value or due to needing to win with Black due to match score. Only rarely do they choose defenses that are worse than -.40. Top engines would rarely choose defenses worse than about -.30 on their own, even with MP randomness. But it is necessary to include openings that are close to your 0.70 line in order to avoid draws. You are in effect just doing a mild version of the UHO idea when you do this, which I think is fine. It's a good compromise between "correct" chess and the need to avoid 100% draws.

But then – isn’t it a bit unclear what these rating lists actually measure?
We have a clear idea when it is about the usual setup – two opponents facing each other, doing their very best from move 1. This experience mainly comes from a long history of human games.
As chess is so drawish, you need a bazillion of games to decide who is the stronger one in a match or tournament between strong chess engines.
So you introduce very uneaven setups with strange and unusual openings. Is it +that+ obvious, that you need the same qualities as in a „normal“ game this way?
Yes, you get clearer answers strength-wise, but how do they translate to the traditional setup if you do enough games?
As engines become ever stronger, the starting positions will become ever more lop-sided this way- at least this is what intuition suggests ( I have no data to prove this).
This might lead to a strange evolution. Human players are obviously mostly interested in „breaking the draw“ while doing analysis, getting unexpected wins from what they perceive as drawish positions. But math-wise it is just as valueable to get „undeserved“ draws. And maybe Stockfish has followed this path for a little too long already.

[ImNotStockfish]

An engine that only gets 50% with normal openings but has +100 Elo with unbalanced openings, touting it as a 100 Elo better engine is strange. Then there is the question of what it looks like in position tests. If the +100 Elo engine is not better in position tests, then the 100 Elo makes absolutely no sense and suggests a distorted playing strength.

For this I have a test suite for 200 games with Ponder ON plus a position test suite, and then I go onto the server and play a few hundred games live.

A chess game is nothing more than a dozen "tests positions". A "test suite" is just a very very very small sample size of what Fishtest and tournamets like CCC and TCEC do, test engines in a lot of positions and the one that is able to solve more of them wins.

Stockfish might not do great at some small-sample selection of positions, but its the best at solving big-sample-size real-world test positions

[mig2004]

Unbalanced openings are perfectly fine if each engine gets to play it with both sides against the same opponents. A better engine might convert the advantage into a win or hold the disadvantage to a draw, compared to drawing with advantage and losing with disadvantage. It just shouldn't be so unbalanced that the results are obvious, such as starting with a piece handicap.

That´s my consideration for using UHO´S. And they are not ¨contrived¨ opening lines. They are human played lines in the past by FIDE RATED masters with 2400 elo or more.

[Graham Banks]

Unbalanced openings are perfectly fine if each engine gets to play it with both sides against the same opponents. A better engine might convert the advantage into a win or hold the disadvantage to a draw, compared to drawing with advantage and losing with disadvantage. It just shouldn't be so unbalanced that the results are obvious, such as starting with a piece handicap.

That´s my consideration for using UHO´S. And they are not ¨contrived¨ opening lines. They are human played lines in the past by FIDE RATED masters with 2400 elo or more.

They're contrived to achieve decisive results, at the expense of having fair opening lines.