How to use prediction model from Classification Learner App in Simulink? ("Dot notation not allowed")

I exported a prediction model "PredictionModel" from the Classification Learner App. I want to use this model in Simulink - either in a Matlab Function block or in a Matlab Function box in Stateflow.

In the command line window it works perfectly like this:

>> Maxtemp=6

Maxtemp =

     6

>> Precipitation=3

Precipitation =

     3

>> T=table(Maxtemp,Precipitation,'VariableNames',{'modeTmax','modePrecip'})

T =

1×2 table

    modeTmax    modePrecip
    ________    __________

    6           3

>> y=PredictionModel.predictFcn(T)

y =

     2

So this is one of the code varieties I tried for a Matlab Function box in Stateflow (without the % sign that seems needed for formatting in this forum):

%
function pred=Prediction(Tmax,Precip)
coder.extrinsic('table');
coder.extrinsic('PredictionModel');
coder.extrinsic('PredictionModel.predictFcn');
T=table(Tmax,Precip,'VariableNames',{'modeTmax','modePrecip'});
pred=PredictionModel.predictFcn(T);
end

but apparently I cannot call predictFcn this way, because I'm getting an error message "Dot notation on function call return value is not allowed." Using 'pred=predictFcn(T)'instead; doesn't help either (-> "Undefined function or variable 'predictFcn'").

The same problems when trying to use the prediction model in a Matlab Function block in Simulink.

Is there a workaround for this?

[Wilfred]

I found this possible solution, based on which I simplified to

 % 
function pred=Prediction(Tmax,Precip)
coder.extrinsic('PredictionModel');
pred=zeros(1,2);
pred=PredictionModel(Tmax,Precip);
end %

Using this, I am getting no compilation errors any more, but the following message running the simulation:

"An error occurred while running the simulation and the simulation was terminated

Caused by: Undefined function 'PredictionModel' for input arguments of type 'xxxxx'."

with xxxxx whatever type I assign to Tmax and Precip (double, single, int8, etc.).

Suggestions how to make this work are still appreciated

Answer 1

[Arie Weeren]

Dear Wilfred,

In order to use the prediction model that you created using the classification learner, you have to save the model that is stored in the PredictionModel. For instance, if it is a Discriminant model, you would use

    saveCompactModel(PredictionModel.ClassificationDiscriminant,'ClassificationLearnerModel');

Then, you can use it in Simulink like

function pred = Prediction(Tmax,Precip)
persistent mdl;

if isempty(mdl)
   mdl = loadCompactModel('ClassificationLearnerModel');
end
pred = predict(mdl,[Tmax Precip]);
end