computer chess project

[tomlynch]

Seek fastest ways to have IPPOLIT program create book. Irregular openings, ideal would be for accomplished programmer to do comprehensive work with statistics. Nine years full-time work using Centurion, Fritz, Rybka with 1) g3 including ... d5 2) Bg2 e5 3) e3 Nf6 4) Ne2 , Pirc, and Dutch. Present project structure: Connect Rybka program to Firebird/Quad Core computer yet to purchase and assemble. Write software enabling automation of Firebird, loading it daily with sets of opening positions on hand in pen/paper format. Need to know how to download Firebird/GUI and whatever else needed to operate it. Looking for ideas, info, contacts. Have considerable knowledge and experience with these openings having been (C) player ICC and playchess.com. Currently playing live games. :?:

[jshriver]

Not entirely sure what you're wanting but this sounds like my Goldentree project.

http://olympuschess.com/goldentree/

I did get it to stable, but ultimately abandoned the project because I felt like it was semi-cheating. In concept of having a strong opening tree is fine but I felt like using strong engines to compute these "answers" kinda tainted the pureness.

May reawaken it once my engine is done and will use it for evaluating positions and rebuilding the tree from scratch using data from my 110gig PGN dataset as a seed.

Just my $0.02

[Dann Corbit]

For some chess positions (e.g. gambit openings) you need a very strong engine (Firebird would probably do nicely) and you must analyze 24 hours per each EPD position on a high end machine.
Even then, you might not get the right answer.

You need also to have a statistics database with won/loss/draw numbers for high-end opponents to know if your engine has the right clue.

For instance, you might have engine analysis that says you are up a full piece. And yet the position loses 60% of the games from that point forward. This is a clue to tell you that the engine does not understand the position.

[jshriver]

Interesting. I might have to fork my own code then The current version used crafty as the analyzer.

The gt_client was xboard compliant and was almost done adding UCI support when I canceled it but I could use polyglot.

This is one reason I made that pgn analysis app so I could reduce some of the positions as seeding points from the marcelk dataset.

http://olympuschess.com/strength/

[tomlynch]

Hello and best of luck with your particular project. My goals are clearly stated so I will hold off on specific responses given that no specific question has been raised. To be sure I lack tech experience and have merely pieced together data from prolonged communications. Can't be sure of really anything in this save the results I have logged from own research. Appreciate participation in talkchess.com.