[Computer-go] Fwd: News on Tromp-Cook ?

[Olivier Teytaud]

> These days, programs of Crazy Stone thread, like Zen, Erica, Aya...etc, do
> stochastic simulation, different with Mogo-type, fixed-sequence simulation.
>

I think CrazyStone's update formula allows more flexibility than the
pattern-based approach of MoGo. So it has better potential;
I have no idea of whether one can really benefit from this flexibility, but
there is a nice flexibility, without big computational cost,
and this representation (by marginal probabilities) is a good algorithmic
choice in CrazyStone and related programs. On the other hand, the
probability involved in the Monte-Carlo part, as originally defined in
CrazyStone, were not very efficient, in spite of the fact that they play Go
better (still the apparent contradiction that a Monte-Carlo which plays well
is not necessarily a good part of a MCTS). The modified versions were, I
think, influenced by MoGo's patterns as defined by Yizao (I think other
persons helped a lot for this, for the experimental setup and the
organization of the code - I consider Sylvain as important as Yizao for this
even if Sylvain did not design the patterns - I only consider myself of very
moderate influence :-) ).

We could win a lot on the Monte-Carlo part with "fillboard" and "nakade" and
a few handcrafted patterns. No idea of whether better
formula could be designed using the more flexible formalism of
CrazyStone-like formula.



> Can you point out where did Sylvain claimed the idea of "balancing" of
> simulation in his thesis (
> http://www.lri.fr/~gelly/paper/SylvainGellyThesis.pdf<http://www.lri.fr/%7Egelly/paper/SylvainGellyThesis.pdf>)?
> I check the thesis, Sylvain talked about balancing of exploration and
> exploitation, but not simulation (maybe he was not using such wording
> "balancing"?). David also didn't cite his thesis in his simulation
> balanacing paper. If Sylvain really proposed such concept, I would like to
> cite it in my thesis.
>
>

Sylvain posted in Computer-Go an explanation of that a long time ago - but
without the word "balancing", probably. See page 154.

Best regards,
Olivier
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20110102/e103689b/attachment.html>