How can I generate C code from a Simulink model with a Neural Nework in the Matlab base workspace?

I have a Python Neural Network model, developped with Keras, and I want to embed it in a dSpace controller which works with C code. There is a dSpace Toolbox that imports a Simulink model in the controller by compiling it in C code, the 'Build Model' function.

I have succesfully imported my trained Python Neural Network in the Matlab Worspace with the Deep Learning Toolbox and the importKerasNetwork function. In Simulink I can use the Neural Network for prediction via the MATLAB function block with the following code and simulation of the Simulink model works :

%%%%%%%%%%%%%%

function y = fcn(X)

coder.extrinsic('evalin');

coder.extrinsic('predict');

NN = evalin('base','NN');

y = double(predict(NN,X));

%%%%%%%%%%%%%%

Here is my problem :

When I try to embed my Simulink model in the dSpace controller it doesn't work as :

• coder.extrinsic is not compatible with code generation,
• I am trying to embed a NN that is stored in the Matlab base workspace.

Should I rebuild my Neural Network in Simulink with the Simulink Deep Learning Toolbox blocks or is there a way to solve my problem in another way?

Thank you in advance for your time!

Answer 1

[Jyotish Kumar]

Hi,

The code generator does not produce code for the body of the extrinsic function and instead uses the MATLAB® engine to execute the call. This functionality is available only when the MATLAB engine is available in MEX functions, Simulink® simulations, or during calls at compile time. For more information refer the link below

https://www.mathworks.com/matlabcentral/answers/13190-coder-extrinsic

[johnnynoc4sh]

Hi Kumar,

are there any options to integrate Keras / Tensorflow models into codegeneratable Simulink models? We are using a Speedgoat platform and I would use it for forecasting timeseries.

Best