How to manually implement of Feed Forward Neural Network processing functions?

I am trying to implement a feed forward neural network created by Matlab nftool on Arduino board.

When I did that all results were incorrect.

I recognized that I did not include preprocessing functions in the implementation so I decided to try to implement the network in Matlab M file first before implementation on Arduino to make sure of my work.

The neural network script contains the following two lines of codes:

net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'};

net.outputs{2}.processFcns = {'removeconstantrows','mapminmax'};

I changed those lines to the following:

net.inputs{1}.processFcns = {};

net.outputs{2}.processFcns = {};

to have a network without Pre/Post-Processing Functions.

In training I implemented the processing function (mapminmax) for the input and target data using the following equation:

 y = (ymax-ymin)*(x-xmin)/(xmax-xmin) + ymin; where ymax=1 and ymin=-1

I neglected the removeconstantrows processing function.

When I used this network the results were very close but not exactly the same.

So I need to know what is the problem in the implementation of processing function because this is the only part I changed.

Thank you very much in advance.

[Greg Heath]

What does "results" mean ???

Please be very specific so that this doesn't have to be an unnecessarily long interactive post.

If you have to, post relevant code.

Greg

[Hatem Elattar]

Results mean the Neural network output

Answer 1

[Greg Heath]

Your normalization equation incorrect. There are separate normalization equations for the input x and target t which I'm sure you can derive.

The resulting normalized output yn is unnormalized using tmin and tmax.

You may also want to check the codes using the HELP, DOC and TYPE commands.

Hope this helps.

Thank you for formally accepting my answer

Greg