How to calculate test loss value from a trained CNN network in case of multiclass classification?

I have few pretrained deep networks for multiclass classification. To compare their performance, besides F1 and AUC scores, I want to calculate the final loss values during test time for each network.

Is their a predefined function in matlab to do that. If not, can someone guide me how to do that using predicted test scores?

[Greg Heath]

To make life simple, why not just use 0 and 1 targets, Sigmoid or Gaussian outputs that are bounded to (0,1) ???

help patternnet

doc patternnet

Hope this helps

Greg