nprod.m version 1.3

Replacement for normprod that correctly treats 0, Inf, -Inf in the weights.

A replacement for the transfer function normprod.m in the MATLAB NN toolbox. With this transfer function, the NaNs in the simulation of RBF neural networks can be avoided.

The function can handly zeros, and positive and negative infinite values in the weight vector. Also combinations of positive and negative infinities are. allowed. Formula is (#Inf*max(t)+#-Inf*min(t))/(#Inf+#Inf) where t is the vector to weigh and #Inf resp. #-Inf are the number of positive respective negative weights. If there are no mixed Infs and -Infs then the value is calculated by the limit (i.e. for 0, Inf, -Inf)