Hi Arsal,
I was not able reproduce this issue you mentioned, so if you can provide the precise conditions about when you get the error and when you don't (e.g. code when you get the error vs code when you don't, which MATLAB release, are there other error messages, etc.), it would help. I would suggest reaching out to MathWorks Technical Support team with these details for further assistance.
In any case, the way persistent variables are used is precisely the way you've written; the first assignment is wrapped with an isempty call:
persistent mdl;
if isempty(mdl)
mdl = loadLearnerForCoder('EnsembleModel.mat');
end
So the way you've written in your post is the correct way.
Additionally, the machine learning models saved using loadLearnerForCoder can actually work this way:
function y = predictActivity(u)%#codegen
n = loadLearnerForCoder('EnsembleModel.mat');
y = predict(n,u);
end
i.e., without the persistent variable. In the generated code, this is doing the right thing; the model is initialized using loadLearnerForCoder only once in the code. Thus, for simulation with the MATLAB Function Block and code generation, you don't need the persistent variable initialization.
