For one hidden layer with "tansig","purelin" transfer functions the post process Fcns = {'removeco​nstantrows​','mapminm​ax'} is done after tansig or purlin?

I have one hidden layer with "tansig","purelin" transfer function. I used net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'}; and net.outputs{2}.processFcns = {'removeconstantrows','mapminmax'}; as post processing, This post process is done after "tansig" or after "purelin"?

Answer 1

[Jayanti]

Hi Sanaz,

In MATLAB’s neural network architecture, each layer is assigned only one transfer function. Since you mentioned both 'tansig' and 'purelin'. I am assuming 'tansig' is applied to the hidden layer and 'purelin' is applied to the output layer.

The output postprocessing functions defined are always applied after the final layer of the network which in your case uses the 'purelin' transfer function. So, the postprocessing happens after 'purelin', not after 'tansig'.

Hope this is helpful!