
384 Chapter 20
2)
A
network may already exist which has proven competence
or sentimental value. It might be impossible to convince
the powers-that-be to switch to a probabilistic neural
network. This method allows us to compute confidences
from an existing network.
3) We must reluctantly admit at least the possibility that a
probabilistic neural network may have performance that
is inferior to that of some other model. This is rare, but it
can happen. In this case, a hybrid model may be appropri-
ate.
The hybrid scheme just described is presented in more detail on page
235.
Hypothesis Testing versus Bayes
1
Method
The reader may be thoroughl ...