No normalization applied in a feed forward neural network.

Hi, y2 at the end of the code should actually be a normalized output. It actually value is:

   -0.2535
   -0.3109
    0.9627
    0.9466
   -0.6760
    0.3281
    0.4481
   -0.1759
    1.1334

% SSVM_InputVectors_Transformed.mat
% x -> 6x149
load('SSVM_InputVectors_Transformed.mat', 'x');
% SSVM_TargetVectors_Transformed.mat
% t -> 9x149
load('SSVM_TargetVectors_Transformed.mat', 't');

% net configure.
net = feedforwardnet(10, 'trainscg');
net.inputs{1}.processFcns = { 'removeconstantrows', 'mapminmax' };
net.outputs{2}.processFcns =  { 'removeconstantrows', 'mapminmax' };
net.performFcn = 'crossentropy';
net.performParam.regularization = 0.3;
net.performParam.normalization = 'standard';

% net train.
% includes already preprocessing and postprocessing.
[net, tr] = train(net, x, t);

plotconfusion(  t(:, tr.testInd), net( x(:, tr.testInd) ), 'custom', ...
                t(:, tr.testInd), weakLearn( x(:, tr.testInd) ), 'toolbox' );
%
y2 = net(sample);

[Mudambi Srivatsa]

I understand that you are expecting normalized output from the feed forward network. By normalized output you mean the values of y2 to be in the range of -1 and 1 correct? Is it possible to share the MAT files used for 'x' and 't' values?

I see that the 'performFcn' is set to 'crossentropy'. Did you encounter a warning message that says "performance function set to mean squared error" when you executed the code?