May 26, 2010
Rybka 4 is the latest UCI Rybka chess engine. The changes since Rybka 3 are extensive.
- The evaluation is more accurate and better aware of king safety and other dynamic factors.
- The search is more efficient in every major category of position.
- Some new endgame heuristics have been added.
- The overall engine behavior has been cleaned up and made less quirky.
- New parameters controlling engine output, engine priorities, and engine time management during game play.
- Ability to take advantage of large memory pages (_link_) and SSE4.2 and SSE4a instruction sets.
- New logging capability, which documents in .html the evolution of the main line for all analyzed positions.
- Asymmetrical piece scoring (see below).
Asymmetrical piece scoring
It is now possible to adjust all piece values for both colors in Rybka 4. This functionality has three uses:
1) Rybka can be informed that some pieces in some specific position should or should not be traded.
2) Setting higher overall piece values for one color implements contempt. Rybka will attempt to not trade the more valuable pieces, keeping more tension in the position, and she will give higher evaluation scores to the side with the more valuable pieces and therefore avoid draws by repetition.
3) Piece values which are more consistent with top GM practice can be used. For still-unknown reasons, rooks and queens tend to be more valuable in computer play than what top human players believe to be correct. The default values are optimized for computer play. GM Larry Kaufman recommends the following adjustments as more human-like: pawns=0, knights=24, bishops=24, rooks=12, queens=0. (_link_)
Please note that contempt is now off by default and should be set manually when appropriate.
Please see the Rybka 4 user area (_link_) or visit our forum (_link_) for more information.
A huge thanks to everyone on the Rybka team, as well as to everyone else who has contributed in any way.
- Lukas Cimiotti, for systematically pushing our clustering project through all sorts of ups and downs to being the weapon and asset which it is today, and for the comically shocking effect produced by his mountain of hardware.
- Jiri Dufek, for an awesome Rybka 4 book, for his pragmatic and highly successful tournament preparation, and also for some useful networking skills.
- Jeroen Noomen, for the groundbreaking Rybka 3 books, for several years of superb tournament preparation, and for making sure that Hans stays out of trouble.
- Nick Carlin, for his unique scientific brand of book preparation, for delivering several tournament victories in a short but productive period of time, and for returning to make sure that Rybka 4 was done right.
- Felix Kling and Christoph Kling for our excellent web site, for interesting reports, and for keeping the forum free from marauding bandits.
- Larry Kaufman, for his innovative man vs machine matches, for countless suggestions about quantitative aspects of Rybka's evaluation, and for helping with the release and tuning the human material weights.
- Hans van der Zijden, for going anywhere and everywhere to operate and represent Rybka, for proudly wearing whatever "medals" tournament organizers foist on him, and for staying out of trouble.
- Dadi Jonsson, for hosting the Rybka forum - the best forum I have seen on the internet - and giving Felix something to keep free from marauding bandits.
- Steinar H. Gunderson for Microwine, which allows Rybka to run on Linux despite my chaotic coding practices, allowing Felix to remain in the dark about the modern invention that is Windows.
- The Rybka 4 Beta testers for a fun, lively and productive Beta testing period: Majd Ansari, Wieland Belka, Shaun Brewer, Nick Carlin, Lukas Cimiotti, Jiri Dufek, Mathias Feist, Steinar H. Gunderson, Dadi Jonsson, Larry Kaufman, Felix Kling, Joachim Nettelbeck, Jeroen Noomen, Kostas Oreopoulos, Ulysses P., Teemu Pudas, Albert Silver, Shahar Tzafrir, Ernst Walet and Victor Zakharov.
- Mathias Feist, Matthias Wuellenweber and Frederic Friedel from ChessBase, for their organized, professional work, for showing me how a small business should be run, and for putting up with my learning curve on release issues.
- Victor Zakharov, Sergey Abramov and their team from Convekta, for their great enthusiasm, energy and creativity, and for having no fear of the impossible
- Nelson Hernandez, for a professional-quality video interview, complete with interesting background noises, and for provoking me into all sorts of predictions and statements which have so far not come even remotely close to being realized.
- Alan Sassler, for walking me through the arcane issues involved in tuning highly correlated feature sets.
- Shahar Tzafrir, for making sure that our cluster has a tough opponent.
- Rybka forum regulars, for all kinds of ideas, and for making sure I keep my feet on the ground.
- For anyone I may have accidentally left out.
- And, saving the best for last, to Iweta, for being great!
I'll keep the details a surprise, but we'll have a very cool announcement within two weeks.
>and also for some useful networking skills
I'll keep the details a surprise, but we'll have a very cool announcement within two weeks.
hmm, can you at least let us know whether or not it costs us money? :D
> hmm, can you at least let us know whether or not it costs us money? :D
Not at all, the announcement will be complete free. :-D
> Please note that contempt is now off by default and should be set manually when appropriate.
How can contempt be set? (unless you mean "Asymmetrical piece scoring = Contempt", which is unclear).
Edit: Did further testing, anytime i enter a value with a decimal point, for example 1.1, they are changed to -1000.
It does not make sense to have the Q have 0 value, unless those values abovve are changes to certain values. If so, how does one change the unknown values and to what new values?
> It does not make sense to have the Q have 0 value
It means "0 offset from defaults", that's why you can set negative values.
> If so, how does one change the unknown values and to what new values?
You look at the engine's score of a given position and tweak them until Rybka scores them as you want. Larry looked at 20 materially unbalanced theoretical positions to get his values (he didn't need to know the actual piece values, just how pieces appeared relative to each other on Rybka's evaluation scores.)
In the Arena configuration window for Rybka 4, i don't see any way to load a personality file. How exactly to load a personality file?
I suppose something must be saved to the registry, so Arena 'remembers' these settings.
> In Arena, you can change the settings, then click the "Save as New Engine" button. This does the same thing as duplicating an engine and changing the settings.
That's what I meant by "personality".
Vas, please allow me a friendly comment on that, as an admirer of Rybka and as a retired successful small business manager, with all due respect.
I don't think that I am alone in thinking that Chessbase's approach to "release issues" is sad: no customer dialogue or support to speak of, vapid new versions, no updates, etc... To see you going down the route of emulating them is simply disheartening. And let me tell you why is consider it the wrong ethical choice.
Like you, I was fortunate enough to have a gift that put me on top of my (modest) field. Like you, I offered stunning value for the price versus competition. Competitors were calling me, kindly suggesting me to push up my price: "You could sell it for much more. Come on." "You don't have to offer some much for such a price", etc... But at the end of the day, it was between me and my clients, my "fans". Yes I could extract more from them. Moneywise. Well, only some of them: those who would afford the new price point, the new, deteriorated value proposition. So I maintained permanently opened client relationships even when success and fame meant that, honestly, that was not economically justifiable. I have never skimped on the performance offered, only the best allowed by my capabilities at all times, even though the competitive field did not require it. The I-could-do-better-but-my-product-is-good-enough-for-you-to-buy-it-as-is-mr-client approach is just ethically bankrupt. As is a worsened reputation in exchange of more money. It is all about a balance between what you give back to the community thanks to your gift and ... greed. Greed. Your client base senses - correctly I believe - a new, deteriorated value proposition. It senses that greed has gone up several notches. I am saddened that you have apparently taken that decision, under the regrettable influence of Chessbase, but, unlike many, I am not resentful: I respect the freedom of the entrepreneur and his newfangled mercantile cynicism.
I won't buy Rybka 4 though. I can afford it many times over, but you've lost my trust, I'm afraid, exactly like Chessbase did a long time ago with its so-called "releases".
>The I-could-do-better-but-my-product-is-good-enough-for-you-to-buy-it-as-is-mr-client approach is just ethically bankrupt.
Yeah and in your business I bet you never had half of your product stolen everytime you made a new product. You can not compare the business of computer chess to most business.
If you refer to Rybka cloning, and unless I've missed a step, its illegality has never been established or even argued in court. So the presumption of innocence applies.
I am sure as a former small business owner you realize the value of concentrating your limited energies.
I wish you the best of luck with your new rental business model anyway: it is an exciting experiment, even if it possibly sounds like very bad tidings for the impecunious hobbyist.
We'll have some things for the impecunious hobbyists out there too, but it will take a couple of years.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill