Simulating neural network with code

I trained multiple neural networks in Matlab for head pose estimation. To implement the neural networks in a real-time setting I want to simulate the networks in another language, so I start writing code for Matlab first.

When I tested this code on the training set, I found out that my outputs (yaw and pitch angle) are slightly different from the outputs the training gave me. The differences are on average only one degree, but it should exactly be the same.

I already did pre- and postprocessing with mapminmax. Can anyone please tell me what I am missing here!?

My code:

    target_train = target_train';

    [~,outMapP] = mapminmax(target_train(1,:), -1, 1);
    [~,outMapY] = mapminmax(target_train(2,:), -1, 1);

    pitch = [];
    yaw = [];

    for i = 1:1568;
        image = input_train(i,:);

        %Map image between -1 and +1
        [~, inMap] = mapminmax(image, -1, 1);
        image = mapminmax('apply', image, inMap);

        adder1 = image*IW';
        layer1 = tansig(adder1'+b1);

        adder2 = LW*layer1;
        layer2 = purelin(adder2+b2);

        out_pitch = layer2(1,1);
        out_yaw = layer2(2,1);

        pitch = [pitch; out_pitch]; 
        yaw = [yaw; out_yaw];

    end

    %Remap the output to the right range
    pitch = mapminmax('reverse', pitch, outMapP);
    yaw = mapminmax('reverse', yaw, outMapY);

[Greg Heath]

MATLAB was invented to be a MATRIX based language to omit unnecessary loops.

Try using matrix equations instead of loops.

Greg