Neural Network in Matlab App Designer

Is it possible to incorporate a Neural Network into an app created from the Matlab App Designer?

[Juan Carlos Pozuelos Buezo]

Hi, i am trying to charge a neuronal network (net.mat) into the app designer. The net is for get a forecast of electricity load. The input is a table with predictors. How i can call de net into de app designer?

For get the result: have i to use function “sim”?

[ypredict]=sim(app.net,input);

I try but i have problem

Thanks fotmr your help, and sorry for my English

[Kojiro Saito]

Which neural network are you using?

For example, in order to forecast numeric values, you can use LSTM networks for regression.

https://www.mathworks.com/help/deeplearning/examples/sequence-to-sequence-regression-using-deep-learning.html

https://www.mathworks.com/help/deeplearning/ref/predict.html

In this case, you can use predict.

YPred = predict(app.net, input);

If you're using Autoencoder, you can also use predict for forecasting.

https://www.mathworks.com/help/deeplearning/ref/trainautoencoder.html

YPred = predict(app.net, input);

[Juan Carlos Pozuelos Buezo]

Thanks Kojiro. From Deep Neural toolbox i used the “neural network fitting tool” to get a model (pre trained). The model i saved as “net.mat”. When i used the net outside the app designer, i write y_pred = sim(net,input_target) and works, But into the app designer i can’t used that.

I have tried y_pred=Predict(app.net,input_target) but “predict” is not accepted in app designer. I have tried y_pred=net(app.net,input_target), but it wasn’t work too.

Can you help me??

[Juan Carlos Pozuelos Buezo]

Hi.

I wil try use the recomendation.

first, i used starupTcn(app) to load de net model. its call netdia.mat:

properties(Access = private)
        net;
        Y_predict;
 end

function startupFcn(app)
    app.net.net = load("C:\Predictor Demanda S.N.I\net\netdia.mat");
    app.net = app.net.net;
end

then, into de code i tried to call de net with "YPred = predict(app.net, input);", like that:

Y_predict = predict(app.net,Xp_prediccion);

but, i got this error:

Undefined function 'predict' for input arguments of type

'struct'.

Error in

Predictor_Demanda_SNI_Corto_Plazo3/ProcesaDIASIGUIENTEButtonValueChanged

(line 336)

Y_predict =

predict(app.net,Xp_prediccion) % el

modelo previamente entrenado

únicamente generará la salida

Error using matlab.ui.control.internal.controller.ComponentController/executeUserCallback (line 378)

Error while evaluating StateButton PrivateValueChangedFcn.

what is my error?

thanks

[Kojiro Saito]

The structure to be input in predict might be different.

How about changing as the follows?

function startupFcn(app)
    app.net = load("C:\Predictor Demanda S.N.I\net\netdia.mat");
    app.net = app.net.net;
end

If the above does not work, add a breakpoint in the line of

Y_predict = predict(app.net,Xp_prediccion);

and pause at this line. Then, what is a variable type of app.net? You can konw the variable type by whos command.

[Juan Carlos Pozuelos Buezo]

Thanks a lot. It works ok ?

Answer 1

[Kojiro Saito]

Yes, Neural Network (both shallow and deep neural networks) works with applications created by AppDesigner both for training and predict. Could you specify your use case, if possible?

[Justin Primero]

I'm creating a graph through the use of a photorestistor operated by an Arduino. I would then extract numerical voltage values of the graph and record specific points of it in an excel file. The values would then be processed in a trained neural network for it to classify the graph that I generated. Is this possible? If so, how can I call a neural network into the AppDesigner?

[Kojiro Saito]

Thank you for sharing your use case. Yes, it is possible in AppDesigner apps.

Here is one solution.

1. Train a neural network model and save the model as a mat file (for example, net.mat)

2. In AppDesigner, create a Private Property(net).

3. In a startup function, load the mat file and set as app.net.

  net = load('net.mat');
  app.net = net.net;

4. In a function callback which processes graph data (inputData), use classfiy with neural network.

  [YPred, scores] = classify(app.net, inputData);

[Justin Primero]

This has been very helpful. Thanks a lot, Kojiro Saito.

[Fernando Hernandez]

Estoy creando un calisificador de sonidos cardiacos, mediante el modulo max4466 el cual es procesado con filtros analogicos y posteriormente digitalizado para su procesamiento digital. Una vez realizado esto, entrara a una red autoencoder, la cual tendra una base de datos con sonidos cardiacos, y asi determinar en que categoria entra el sonido, si en sanos o no sanos.

Answer 2

[Pablo Salaverria]

Dear all,

while creating an app using app designer I am not able to compeltelly deploy my trainned CNN. My main issue is wiht the function classify which should be valid with:

[YPred]= classify(app.net,app.imds1);

Where app.net is the trainned CNN, app.imds1 is the imageDatastore where all the images that I want to classify are stored.

The error message that I got is:

Error using classify (line 123)

Requires at least three arguments.

Meaning that the function classify is using a 2nd options instead of the one that I want it to use. Is there a way to make it work?

[Kojiro Saito]

I have just answered your question in this link. Please take a look.