So if that holds, in December Stockfish or some other engine is going to overtake Rybka, it seemed likely in June, let's see what happens.
There is an unmentionable engine that I think has pulled just about even and has a really different behavioral style. I consider it the #2 engine right now, and one more bound and it could be #1. Of course, this judgment is subject to the tests some are conducting at different R4 settings.
R4 and the other engine is about the same elo.
P.S. It would be nice to see an update for R4.
I think Tord has again made a very big contribution to "reusable" code with his new remote Glaurung on Ipod code. I hope some of the commercials including Vas will be able to find some ideas there that could be used for remote versions of their software. I think only Tord could do this much, in the limited time that he has for chessprogramming. But the other Stockfish programmers I'm sure are also not sitting still
The code of stockfish1.9 is smaller than the code of stockfish1.8 and it means that it may be easier also to improve the elo of stockfish in the future.
I believe that the evaluation of stockfish1.9 is relatively simple(evaluate.cpp of stockfish1.9 is only 950 lines of codes even when we include all the comments)
and there is a clear room for improvement by a better evaluation.
Edit:The evaluation function is done also in other files and not only in evaluate.cpp but I still think that it does not change the fact that the evaluation of stockfish is relatively simple and there is a room for improvement in it.
Nevertheless, engine vs engine is a very different matter and it is possible that we may see SF getting positive results against R4 in one year or so (i am assuming we will have one SF update every 6 months, and we will not see any R4 update anymore, since Vas has never done what he said)
> Vas has never done what he said
Well, in his defense, that is obviously not true. It is true that he has disappointed us time after time. But he hasn't been a pathological liar or anything like that. His credibility is not the greatest, but not because he is deliberately trying to deceive us. There are different types of bullshitters, and Vas is one of the more benign varieties. He has never hurt any of us.
> for long analysis I still think R4 is much reliable and stable than SF
No argument from me, though R4's well-known endgame evaluation problems are vexing. For me, the engine that most accurately and quickly identifies a drawn position as 0.00 is a treasure. There are others that do this better even though they may be weaker in other respects.
I guess he must have other priorities and that he has full right to do so. I think we all agree that if he says he is going to write R4's FAQ document we can wait and wait. If he says he is going to release 3+ version, we can wait and wait. If he says he is going to release 4.01...
But who cares? At the end R4 with all her bugs is the best engine so far...
> There is no way Rybka is the best engine
According to any objective criteria?
Rybka does not have any negative issues if you consider trivial the mpv bug, lack of support for persistent hash, misevaluation of rooks with pawns endings. BUP and wrong bishop endigames.
Rybka is probably the best in the middlegame, that's why it's a must have if you play correspondence chess but I'm probably not being objective.
It is supposed they must have other bugs, or in case they are bug-free (?), it seems that they are not good enough to beat a buggy engine like R4 at those time controls.
> Vas is one of the more benign varieties
I agree with this assessment. Maybe I'm wrong and he's Mr. Evil in his spare time but I have the impression that Vas is like many other engineers I've known who constantly struggle with time management and lose interest in projects as time goes by. Neglect is a sign of boredom and it's often followed by a new project with a new set of challenges...
The side that said that beancounters would win in the end (Chris Wittington was one of those, I believe.) has completely won that argument, or so it would seem at this juncture. Personally, I think that's a sad development because beancounting isn't nearly as interesting. I'd much rather see developers focus on building new and interesting personalities. Seeing an engine like Stockfish that can reach insane depths when analyzing the starting position and knowing that those depths are meaningless noise does absolutely nothing for me. If Stockfish is the future of computer chess, we're screwed.
> Move choice is a product of evaluation and search, I hate Stockfish's search and its super-instability, but I love Sotckfish's move choice and I think it's its main strength, so as in the other thread I'll say that if Stockfish's evaluation is poor I don't want to see it fixed.
We're in agreement. I'm glad it exists and I use it often in certain types of positions but I fear that in a few years we won't have the variety we have now as the cloners drive Vas into poker or Go or whatever his next project is (Rybka, I fear, is dead) and ideas from Stockfish make their way into other engines. If that happens, it will be very bad for computer chess.
> (Rybka, I fear, is dead)
Nah, he just needs to change his model and damn the logistics, he could rely on people buying the engine monthly or something drastic if necessary, so the cloners are always very behind on his ideas.
But perhaps that's not even necessary, I highly suspect that Rybka 4 is currently selling as good as it would with or without the clones, just like Rybka 3 was selling well in the 6 months before Rybka 4's release (it's hearsay, but I think the reason Vas switched to a 18month release model, if the engine sells as good as new, why release a new engine so soon?)
That would be a great product.
Well, I'm still hoping for Persistent Hash support, but if 4.1 doesn't get it back, it's probably gone forever...
I do not agree.
I prefer to see stockfish's evaluation to be fixed first.
I want a program that static evaluation of some high number that is not mate means winning when the only unknown question is the number of moves to win(there are positions when I am sure that one side is winning even when I do not know the number of moves to mate and I think that it is better if chess programs also get the knowledge to identify this type of positions for better pruning.
I've previously praised Stockfish for its overoptimism, suggesting a move and saying it's 1.00 (and the only one that high) while most other engines think the top 4 moves are equal, and Stockfish being right that the 1.00 mover gives white the best chances. So the 1.00 is perhaps too high, but I don't take it as "this move wins the game" but as "I really like this move", one just needs to get used to Stockfish scores even if they're meaningless (as in, it could be 0.40 or 1.60, and it will most probably jump around), what I meant to say is that if "fixing" Stockfish's evaluation will get rid of those jumpy moves (and if Stockfish will say the top 4 moves are equal like the other engines) I don't want to see it fixed.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill