How to integrate trained neural network in the app (I have referred to older posts but still not succeed).

Dear Matlab Users,

I have refered to several answers by @kojiro Saito on how to integrate trained network in the app designer. One of the link that I refer is

https://www.mathworks.com/matlabcentral/answers/508824-how-do-i-integrate-a-trained-neural-network-into-an-application?s_tid=srchtitle

and I have followed his advice, but somehow the codes still not working. I hope someone could point out the mistakes that I made or how should I improve the codes.

first I add properties

  properties (Access = private)
        Property % Description
        net;
        ypredict;
    end

then I add new function to load the network

 function  startupfunc(app)
         net=load('net.mat');
         app.net=net.net;
end

however here, received an error of Line 33: A METHODS block or END might be missing before the function definition. This might be causing additional error messages.

Next I tried to predict using the trained network by 4 variables input by the user.When calculate button is pressed, the new output shall be shown on the interface but no values appeared.

 % Button pushed function: CALCULATEButton
        function CALCULATEButtonPushed(app, event)
            T=[app.Variable1.Value,app.Variable2.Value,app.Variable3.Value,app.Variable4.Value];
            [T1 ps]=mapstd(T);% my network was trained in mapstd 
            ypredict=predict(app.net,T1);
            yreverse=mapstd('reverse',ypredict,ps);
            app.Output.Value=yreverse;
        end

your kind assistance are much appreciated.

Answer 1

[UDAYA PEDDIRAJU]

Hi Nurliana,

I understand that you’re trying to integrate the trained neural network in the app designer. It seems that the error you encountered might be due to the function definition within the wrong section or a missing method block. To integrate a trained neural network in the app, you can follow these steps:

1). Load the Trained Network: Ensure that the function to load the network is within a methods block. You need to encapsulate the function within the "methods" section of your app.

methods (Access = private)
    function loadTrainedNetwork(app)
        net = load('net.mat');
        app.net = net.net;
    end
end    

2). Predict Using the Trained Network: When the "Calculate" button is pressed, ensure that the input is pre-processed as per the training data, and then use the trained network to predict the output.

             
methods (Access = private)
    
    function calculateButtonPushed(app, event)
    
        T = [app.Variable1.Value, app.Variable2.Value, app.Variable3.Value, app.Variable4.Value];
    
        [T1, ps] = mapstd(T); % Assuming the network was trained using mapstd
    
        yPredict = app.net(T1);
    
        yReverse = mapstd('reverse', yPredict, ps);
    
        app.Output.Value = yReverse;
    
    end
    
end   

By encapsulating the load function within the methods block and ensuring proper preprocessing and prediction steps in the “calculateButtonPushed” function, you should be able to integrate the trained neural network into your app successfully.

You can also refer to the https://www.mathworks.com/matlabcentral/answers/444474-neural-network-in-matlab-app-designer.

Hope this helps!