If prior probabilities are unknown, I suggest MinMax rule as described in Hand (1981) page 7.
The Bayesian decision rule, in MinMax approach, is designed so that we minimize the maximum possible risk.

Hand, D. J. (1981), Discrimination and Classification, John Wiley.


A 21:29 23/06/2003, qhwang a écrit :
>Hi folks,
>
>Can anyone explain me about how to determine the priori probabilities when Bayesian decision rule is applied, say, to classify a pattern into object class and non-object class? I know usually equal priori is assumed, but in this case how can the training set for both classes be interpreted? I mean, are these priori probabilities not necessarily derived from the training set since usually in which postive samples are not equal to negative samples?  Any help will be greatly appreciated.
>
>Many Thanks.