# How should three unordered categories be encoded in a bayesian network framework?

The SAS FAQ suggest that for unordered two categories I should one dummy variables, for example:

The common practice of using target values of .1 and .9 instead of 0 and 1 prevents the outputs of the network from being directly interpretable as posterior probabilities, although it is easy to rescale the outputs to produce probabilities (Hampshire and Pearlmutter, 1991, figure 3).

Following above logic, how do I know which target values I should use for three categories and why?