RomiChess && learning or the emperor has no clothes

[Chan Rasjid]

Hello Michael,

Some things can be understood with a faily high degree of certainty whether it works or is of value just from the analyical perspective. Some things cannot. But the person studying the matter it must be qualified. It would be difficult to trust someone who has not written a chess program to make judement if your learning works.

Take the eternal debate search-versus- evaluation - it will always remain only as a point for debate. Take, late-move-reduction, reasoning + testing will not confirme IF IT IS GOOD , it may not work for everyone with positive result. Nullmove is a certainty through empirical findings and has a simple principle behind it.

I do chess programming but never ever think about book learning.
My instinctive feeling is probably with Dr. Hyatt. I believe it can be understood from just the analytical perspective and no testing or experiments needed.

Best Regards
Rasjid

[Michael Sherwin]

The human opponent that I had in mind most is the club or tournament player that wants an interesting 'tool' that they could use to hone a particular line of play for use against other humans. So lets say that someone would like to prepare the dragon variation for compitition. If they start off by loosing then Romi will play back her winning line as book each time untill they vary thier play to find a way to win, and when they do find a way to win then the shoe is on the other foot and it is Romi that will start varying her play untill she finds a way to win. This sounds like a good way to go about trying to make an interesting program that some people would like to have. Would you at least give me this much?

[Dann Corbit]

Certianly, it would be useful for working on AN opening.

Suppose, for instance, that you wanted to pound the spots out of the Sicilian Dragon. You could set up that start position and let your program learn away at slow time control against a raft of opponents. Eventually, you would probably come up with some interesting novelties.

Instead of deciding now what it is and isn't good for, why not try it out?

When you give your program away, it also affords an opportunity to have lots of people run the program so that you can accumulate a large learning file if they all send their files back to you and if you have a way to merge them.

I don't think you should think about how it can't be used. Think about how the idea can be used. Certainly, it is good for something.

[Michael Sherwin]

Thanks Dann. I am very close to making a new release!

P.S. Yes, Romi can merge pgn files into her learn file.

[Chan Rasjid]

So if my guess that your matter can be understood in some way from the analytical perspective is correct, then now I stand on your side.
You know what you want to achieve and what your book learning feature can do. Even if only just a few are interested (probably not Dr. Hyatt), it is completely legitimate.

Life's only meaning is legitimate interest.

Rasjid

[Michael Sherwin]

[Michael Sherwin]

RomiChess p3h with learning fixed. Download at http://wbec-ridderkerk.nl/html/enginesindex.htm. To use the learning feature as a human player I know that Winboard will handle that propperly, Arena 1.08 does not seem to work. I imagine that there are other gui's that will also work correctly.