I have used the train function of a net built via the newpr function in order to obtain a neural network (the network has 2 inputs, 2 hidden layers with 4 neurons each, and 2 outputs). Now, I would like to know all the characteristic of the final net. I am able to see the 4x2, 4x4, and 2x4 matrices corresponding to the links between neurons. I am also able to see the 4x1, 4x1, and 2x1 matrices corresponding to the biases. Since I am using the newpr function, I assume that the Logsig or the Tansig function is used at each (non input) neuron. However, if I simulate the network with all these parameters (even, for instance, in an Excel spreadsheet) I do not obtain the same results of MatLab with the same inputs. Is there something I am missing? For instance, could it be that the input are pre-processed and the output are post-processed? Thank you for any help!
