Huh, learning, what is it good for? Absolutly ...

[Michael Sherwin]

So many things to try, I do not know what to suggest instead of what you are already doing. The Grunfeld is not a very good choice for Romi as you have noted. OTOH, to see if Romi eventually learns her way to good chances in the Grunfeld is very interesting.

Romi's learning is most effective when she starts off with a clean learn file and learns her openings from other programs. If you do not tire of this experiment then you may also consider running that next.

Is LL private? If not then I could also run some test here on an athlon 3700+ notebook.

[BubbaTough]

LL is a private, ugly little program. No one but me currently has a copy. I am probably going to clean it up a little (support dynamic hash size reallocation, non-hardcoded tablebase location, and such for example) so I can at least give a non-learning version to a tester or something.

The learning requires an additional thread be running, so the test match I am running is using 3 processors (1 for Romi, 1 for LL, 1 for the the LL learning program). This means most tournament/rating list organizers can't really test it with learning on. It also gives LL an unfair advantage when you compare it with other learning programs.

I am stopping the match for now I think. The match was for useful for me anyway, since it helped me uncover a bug in my learning that I think was responsible for my last round loss to NAUM in the ACCA championships.

Next time I get the urge to do another match I will try your suggestion (just erase the learn file?) and probably use a slightly faster time control. Or if you Tony prefer me to use a different starting learn file just let me know.

-Sam

final result:

LearningLemming 53.5
RomiChess 37.5

[Michael Sherwin]

Happy bug hunting!