In any case, later versions of MoGo used (still use?) an exploration constant
of zero, which means that the UCB formula is no longer used.
Still use. But I'd like to check that it's a good idea, in particular
for the parallel version.
For example, most of our openings are useless for the parallel
algorithm: the parallel algorithm is too much
faster than the sequential one to benefit from opening moves generated
with the sequential algorithm, whenever this
generation was performed with a much longer computation time.
Also, I've met people claiming that
- they need a constant >0 for exploration;
- they could not reproduce the improvement in mogo (they have an
improvement, but not as large as
in the Icml paper).
These two elements are perhaps related in that for some unknown reason,
the Rave implementation
in mogo might be particularly efficient. I just don't know why, but I guess
many things are very implementation-dependent.
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
