training multi dimensional features using newff or newpr

Hello there,

I am trying to train my a neural network using newff. My features are muti dimensional, i.e. having 13 columns and 20 rows against one sample. I have 100 samples of input. Is it possible that I train NN with these features without reducing their dimensions? If possible than how? matlab allows one sample per column so I put all 13 feature-columns of one sample together in one column which I suppose is effceting my results.

would anyone suggest anyother way to tackle this problem? thanks in advance.

Answer 1

[Greg Heath]

If the N input vectors are I-dimensional and the N output vectors are O-dimensional the training matrices have sizes

 [ I N ] = size(input)

[ O N ] = size(target)

To design a single hidden layer MLPNN with I-H-O node topology

net = newff(input,target,H);

Hope this helps.

Greg