|
Anyone has used "patternnet" function in the Matlab Neural Network toolbox? I wounder if how the output transfer function works in this function. By definition, the class labels should be like softmax labels (The target data for pattern recognition networks should consist of vectors of all zero values except for a 1 in element i, where i is the class they are to represent). However, the output transfer function by default is "transig" that outputs values between -1 and 1. So, how these outputs are matched with the class labels which have 0 and 1 values? Another thing is that when I set the output transfer function to "logsig", the trained network will have strange outputs, always between 0.5 and 1. And it does not work (classify) well. It is the same when I change the internal transfer functions. In summary, it works well only by its own default parameters. I need a Net trained with the logsig transfer function for the internal layers and whatever works better as the output one. BTW, I am using this function for a two class task. |
