[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [computer-go] An [open] question on game tree search theory
Don Dailey wrote:
Usually at this point somebody will suggest that evaluations should be
of the form "score + confidence estimate", i.e. that evaluations should
consist of two numbers, evaluation score, and some form of evaluation of
the confidence in the score. A generalization would be to let
evaluations be probability distributions of evaluation scores.
If you view evaluation more as the probability of winning (where 0.5 == draw)
then evaluation which you have little confidence in could be pushed in the
direction of a draw, reflecting your lack of confidence in the reliability of
the score.
What is a probability of winning?
Probability that the playing algorithm will win against a hypothetical,
fixed opponent mechanism---which is circular, because the evaluation
function affects this probability? And what is the opponent model?
Or probability that the game tree node is a game-theoretic win? What
does this mean? One game node is either win or not, so there is no
stochastic experiment. You need to have a repeated experiment or a
population larger than one object to be able to speak of probabilities.
I'm not being rude. I think this is one of the key points. "Probability
of winning" is often mentioned, but what does it mean? Every node is a
game-theoretic win or not, so no probability here. Results from actual
game play, with our imperfect algorithms, depend on the opponents. Is
there an opponent model implied? Against a perfect opponent, all
algorithms that can make mistakes should evaluate the probability of
winning to zero!
Please shew light on this!
Yours Truly,
--
Antti Huima (Mr.)
Director, Conformiq Tools
mobile: +358 40 528 8667
email: antti.huima@xxxxxxxxxxxxxxxxx
Conformiq Software Ltd.
Stella Terra, Lars Sonckin kaari 16
FIN-02600 Espoo, Finland
tel: +358 10 286 6300
fax: +358 10 286 6309
http://www.conformiq.com/
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/