An actual interesting computer chess read about FF2

[dkappe]

In all the lighthearted frivolity about the law, it seems we’ve missed an article on how an ICCF GM evaluates an analysis partner and comes to prefer FF2. Warning: contains actual chess.

https://en.chessbase.com/post/discussin ... al-network

[glennsamuel32]

In all the lighthearted frivolity about the law, it seems we’ve missed an article on how an ICCF GM evaluates an analysis partner and comes to prefer FF2. Warning: contains actual chess.

https://en.chessbase.com/post/discussin ... al-network

And I miss amusing posts

Thanks !!

[towforce]

In all the lighthearted frivolity about the law, it seems we’ve missed an article on how an ICCF GM evaluates an analysis partner and comes to prefer FF2. Warning: contains actual chess.

https://en.chessbase.com/post/discussin ... al-network

Good read - thanks for the link.

One nagging little thought: if it's better at suggesting good moves, then why doesn't it play better chess? Why isn't at the top of the independent rating lists?

[dkappe]

Good read - thanks for the link.

One nagging little thought: if it's better at suggesting good moves, then why doesn't it play better chess? Why isn't at the top of the independent rating lists?

Are any of those rankings outside the margin of error? Regardless, they’re fairly close.

I think one clue is that the GM found that SF gave lots of 0.00 evals, while FF2 ventured an actual opinion. But honestly, I can’t do better than that article.

[Madeleine Birchfield]

In all the lighthearted frivolity about the law, it seems we’ve missed an article on how an ICCF GM evaluates an analysis partner and comes to prefer FF2. Warning: contains actual chess.

https://en.chessbase.com/post/discussin ... al-network

Good read - thanks for the link.

One nagging little thought: if it's better at suggesting good moves, then why doesn't it play better chess? Why isn't at the top of the independent rating lists?

Because of contempt. Stockfish is primarily tested against weaker opponents in the rating lists, for which its default contempt value of 24 centipawns helps defeat weaker opponents, as higher contempt causes Stockfish to complicate the position which weaker engines will invariably blunder in. However, contempt introduces a bias into Stockfish's evaluation, resulting in default Stockfish selecting suboptimal moves at times; this is most evident when Stockfish goes against stronger engines than itself, where having a large contempt like the default 24 is a bad thing and causes Stockfish to blunder.

This has always been one of my complaints about the Stockfish project, that its developers are chasing elo gains on the rating lists and on fishtest at the cost of accuracy by setting default contempt to a value different from zero.

The Fat Fritz 2 compile of Stockfish has the default contempt set to zero, resulting in higher quality and less biased evaluations, but on rating lists weaker engines have an easier time drawing Fat Fritz 2 due to contempt being equal to zero. But Fat Fritz 2 isn't the only Stockfish derivative to have contempt zero, syzygy's CFish also has a default contempt of zero, and if syzygy has updated CFish to be par with latest Stockfish dev, then it should show higher quality move choices than latest Stockfish dev and be free, as opposed to the costly Fat Fritz 2.

[connor_mcmonigle]

Why does the article's author come to the conclusion that Stockfish13 with the FF2 network is superior for analysis compared to Stockfish13 with the default network despite the FF2 network testing weaker than the default network on all independent rating lists?

Well, for one, "suggesting better moves for analysis" is utterly subjective. Suggesting a more trying move (Rab1 instead of Rfb1) in one test position is far from conclusive. Also, it's not like ChessBase is going to publish an article claiming the FF2 network is inferior for analysis. ChessBase, the seller of the product, is the last entity we should expect an honest comparison from.

More importantly, simtests have been performed for the FF2 network showing a >60% similarity between Stockfish13+FF2 and Stockfish13. While simtests are imperfect, such a high similarity score suggests that, for analysis, Stockfish13+FF2 and Stockfish13 would likely be very difficult for any human to even distinguish.

[Dann Corbit]

I doubt if anyone in the world is more qualified than Stephen to offer an opinion.

I think part of it lies in offering moves he likes. I happen to know that Stephen likes aggressive, enterprising lines that offer lots of winning chances. Now, this sort of line might score the same as a line that does not offer the same volume of winning chances.

So perhaps it is a style match with his that he likes.

Perhaps FF2 is tuned with longer time control. I don't know the answer to that. I would be interested in specifics as to how exactly he tuned his network. However, the environment here is a bit hostile, so I suspect he won't be offering explanations in this forum

[connor_mcmonigle]

I doubt if anyone in the world is more qualified than Stephen to offer an opinion.

I think part of it lies in offering moves he likes. I happen to know that Stephen likes aggressive, enterprising lines that offer lots of winning chances. Now, this sort of line might score the same as a line that does not offer the same volume of winning chances.

So perhaps it is a style match with his that he likes.

Perhaps FF2 is tuned with longer time control. I don't know the answer to that. I would be interested in specifics as to how exactly he tuned his network. However, the environment here is a bit hostile, so I suspect he won't be offering explanations in this forum

I should clarify that my post was not accusing the author of lying. It's entirely possible, likely even, that he prefers the FF2 network for analysis. My point was that, given that this article was published by ChessBase and that ChessBase wouldn't publish an article by an individual claiming FF2 to be inferior for analysis, there's certainly some selection bias here.

[Dann Corbit]

Goes without saying.
For years chess companies have put "Word Champion!" on the box when their program won the ICGA tournament.
And they are somewhere below #10 in CCRL and CEGT, bot nobody plays in ICGA anymore except commercial programs with a few Europeans sprinkled in

I don't have any qualms against trying to make money with chess software. But I do wish that chess companies were fully truthful in their ads.
OK, Pollyanna wish.

Now, some might say, "But the ICGA is the world championship!"
Not, in my opinion, since TCEC started.

Now, if you win TCEC, put that on your box and smile from ear to ear.

[gonzochess75]

Goes without saying.
For years chess companies have put "Word Champion!" on the box when their program won the ICGA tournament.
And they are somewhere below #10 in CCRL and CEGT, bot nobody plays in ICGA anymore except commercial programs with a few Europeans sprinkled in

I don't have any qualms against trying to make money with chess software. But I do wish that chess companies were fully truthful in their ads.
OK, Pollyanna wish.

Now, some might say, "But the ICGA is the world championship!"
Not, in my opinion, since TCEC started.

Now, if you win TCEC, put that on your box and smile from ear to ear.

Right, but "actual interesting"