How to eliminate, function not found warnings when using Matlab Runtime

I have built a .net library using Matlab R2018b, which uses a Classification Learner trained model of type SVM (CompactClassificaitonECOC). While using the classification from within Matlab works fine, when I use the .net library, I get all sorts of warnings for it, which say "Could not find appropriate function on path loading function handle..." e.g.

Warning: Could not find appropriate function on path loading function handle C:\Program Files\MATLAB\R2018b\toolbox\stats\mlearnapp\+mlearnapp\+internal\+model\DatasetSpecification.m>@(x)exportableModel.predictFcn(predictorExtractionFcn(x))

Warning: Could not find appropriate function on path loading function handle C:\Program Files\MATLAB\R2018b\toolbox\stats\mlearnapp\+mlearnapp\+internal\+model\+transformation\TrainedManualFeatureSelection.m>@(x)decoratedPredictFunction(featureSelectionFunction(x))

Warning: Could not find appropriate function on path loading function handle C:\Program Files\MATLAB\R2018b\toolbox\stats\mlearnapp\+mlearnapp\+internal\+model\DatasetSpecification.m>@(t)t(:,predictorNames)

> In setupClassification>@(x)exportableModel.predictFcn(predictorExtractionFcn(x))

and so on..

Please help in resolving this. I am using Matlab Runtime 2018b as well.

Answer 1

[Steven Lord]

From your description it sounds like you're loading an instance of an object from a MAT-file in your application.

If you're loading an object into your standalone application without giving any indication in the code of the application itself that the definition of the object will need to be included in the application, the dependency analysis may not include it in the application. [See the Tip in the "Fixing Callback Problems: Missing Functions" section on this documentation page, which though it's for the most recent release I believe is still applicable for older releases.]

You're probably going to need to use the %#function pragma or the -a option to include the definition of the class in your application, or you'd need to indicate in the code itself (say by calling the class constructor) something that informs the dependency analysis routine that it needs to include the class definition.

[Sameer Raheja]

Thanks a bunch for your response. You are hitting the right spot. I do have a trained model which was generated using Classification Learner, and saved as a MAT file. I am loading this MAT file in my application, i.e.

function [imageClassificationModel] = setupClassification()

%#function dispostionModel.mat

model = load('dispositionModel.mat');

imageClassificationModel = model.trainedModel;

end

However, as you can see I have included a %#function pragma here, and am also explicitly passing additional parameters to MCC, i.e. -a <path>\dispositionModel.mat, based on your suggestion. However it has not alleviated the issue. Am I doing something wrong here?

Or do I have to include/embedd some definition or file path when generating the MAT file from classification learner.

[Steven Lord]

A MAT-file is not a function. You need to tell MATLAB that it needs to include the definition of the class in the application, so when it loads the object from the MAT-file MATLAB knows how to create it.

If you tell me you're bringing a dessert (the object) to a potluck dinner, you should tell me (MATLAB, mcc) that it needs to be kept cold before you bring it (when you compile the application) so I can make sure I have ice packs or a small refrigerator (the class definition) available. If not, at best we end up with ice cream soup. At worst, we have to throw it away lest everyone ends up with food poisoning.

What's the class of model.trainedModel? Use that class name in your pragma. [You may also need to specify the name of the MAT-file in your call to mcc, I'm not certain of that off the top of my head. Regardless it shouldn't hurt.]

[Sameer Raheja]

Understood. So I was doing the pragma include you suggested (i.e. ClassificationEnsemble) earlier, and I forgot that I had changed my TrainedModel to ClassificationSVM and missed changing the pragma altogether. However even after doing a %#function ClassificationSVM, and including the path to its definition in mcc, I was still getting the warnings.

I found the other part of solution at another answer thread: https://www.mathworks.com/matlabcentral/answers/354059-deploy-a-model-created-by-classification-learner-as-an-excel-add-in

Apparently, one cannot use predictFcn in such cases of exportable TrainedModels, which is what the Matlab warning is about when loading such TrainedModels, i.e.

Warning: Could not find appropriate function on path loading function handle C:\Program Files\MATLAB\R2018b\toolbox\stats\mlearnapp\+mlearnapp\+internal\+model\DatasetSpecification.m>@(x)exportableModel.predictFcn(predictorExtractionFcn(x))

And no pragma or mcc path includes get rid of this warning.

However, changing my implementation of using this trained model from:

disp = trainedModel.predictFcn(tt);

to:

disp = predict(trainedModel.ClassificationSVM, tt);

got rid of my other warnings. Of course, I had to do a pragma include i.e.:

%#function ClassificationSVM

Thanks for your help.

Answer 2

[Image Analyst]

Some functions cannot be included in a compiled executable. Usually they are applets with a user interface like the Classification Learner, Color Thresholder, Image labeler, etc.. Check the excluded files log file to see what files it cannot include. Call tech support if you still have problems that you think are not related to that.

[Sameer Raheja]

Thanks for the response. Apparently, that does not seem to be the issue. The mccExcludedFiles document is empty and the PackagingLog does not state any such issues. Is there somewhere, I should be looking?

[Image Analyst]

Then I don't know. You'll have to call tech support to solve this one.