How Does Matlab Neural Network Toolbox Preprocess Data?

I try to analyse my neural network trained via Neural Network Toolbox and Matlab 2016a. I was wondering how Matlab preprocesses data. I tried the following: X = -1:.1:1, Y = purelin(2*tansig(X) + tanig(X) -2) and trained a Feedforward Net, lets call it "net", with one hidden Layer and two neurons with tansig activations. I thought net(X(k)) is the same as purelin(net.LW{2,1}*tansig(net.IW{1}*X(k)-net.b{1})-net.b{2}) for any feasible index k, but this is not right. What did I do wrong or is there some hidden preprocessing?

Thanks for any answer!

Answer 1

[Greg Heath]

You forgot the default input and target normalization with MAPMINMAX followed by the output denormalization.

Hope this helps.

Thank you for formally accepting my answer

Greg