MATLAB: Error using Classificationlearner model in Simulink

classification learnercoder.extrinsicfunction handlemodelpredictpredictivesimulink

I used Classification learner App with a Fine KNN model that I called 'model_test'.
Now I want to use it in simulink.
So, I used a function block with the inputs of my model and 'model_test' as a parameter :
function y = predic(Fflown,N1n1,N2n1,N3n1,Taun1, model_test)
coder.extrinsic('model_test');
coder.extrinsic('model_test.predictFcn');
T=[Fflown,N1n1,N2n1,N3n1,Taun1];
y=0;
y = model_test.predictFcn(T);
end
But I run the simulink model, I have the error :
'MATLAB class 'function_handle' found at 'model_test.predictFcn' is unsupported.
Parameter 'model_test''
I saw that I may should use save CompactModel to export my Classification learner but when I try it I have another error :
"Undefined function 'toStruct' for input arguments of type 'ClassificationKNN'.
Error in saveCompactModel (line 17)
classificationStruct = toStruct(classificationObj); %#ok<NASGU>""

Best Answer

Unfortunately this workflow is more painful than it should be. I've found the best set of steps is to:
  • Generate the code for the model in classification learner.
  • Copy out the relevant parts that call the classification training algorithm into a script, in this case fitcknn, and then retrain the model from the script.
  • The output will now be a ClassificationKNN model that you can call saveCompactModel on.
  • Then from the Simulink function:
function c = useModel(x)
persistent model
if isempty(model)
    model = loadCompactModel('savedCompactKNN.mat');
end
c = predict(model, x)
end
You'll notice that I don't need coder.extrinsic as this will generate code.
If you did want to reuse the already trained model, you'll need to grab that property of the trainedClassifier struct and save it:
saveCompactModel(trainedClassifier.ClassificationKNN) % or similar
But then you'll need to map all of the inputs to it which may not be as straight forward as just retraining it with the fitcknn fucntion directly.
Feel free to private message me if you'd like me to walk you through this. I'm trying to find users to convince dev this should be easier.
Related Question