Adds the value epsilon, which is defined in section
, to every element of the hidden state transition
matrix, A, the symbol emission probability matrices, p,q and
r, and the initial hidden state probability vector, pi, and
normalizes them to satisfy the stochastic constrains. This method can be used as
an easy way of changing any parameter whose value might have been set to zero
(by a parameter estimation, for example) and setting it to a really small
value.
Usage: fix_full(N, B)
where N is the number of hidden states; and B is the
observation batch size.