Error using classreg.l​earning.Fu​llClassifi​cationRegr​essionMode​l.prepareD​ataCR (line 213) X must be a numeric matrix.

Why do I get this error when running the "Deep Learning Example: Feature Extraction using AlexNet and CIFAR-10 Dataset"?

"Error using classreg.learning.FullClassificationRegressionModel.prepareDataCR (line 213) X must be a numeric matrix."

[Adam Danz]

I've never worked with any component listed in the question and am 100% unfamiliar with all of it. Yet, I'd bet my next paycheck that some input variable is not a numeric matrix but it should be.

[Guillaume]

Melvyn Smith's comment originally posted as an answer moved here:

That much is apparent from the error code itself. However, this is MATLAB demonstration code (with the images downloaded by the code itself, so I assume they are OK) downloaded from the MATLAB website, here:

https://uk.mathworks.com/videos/using-feature-extraction-with-neural-networks-in-matlab-1492009542601.html

So, why the error?

I am relatively new to deep learning and in the process of deciding whether to use MATLAB for my deep learning needs. I had thought there would be lots of support for it but other than a few introductory videos on their website, there does not seem to be a lot.

[Guillaume]

I don't have the toolbox so can't test the example.

The obvious question is what is X then? Set matlab to break into the debugger when it encounters an error with:

dbstop if error

run the example the code and when it breaks into the debugger, look at the class and content of X.

[Adam Danz]

I googled the demo you mentioned because no link was provided. It led to a file exchange by Math Works and someone else has recently left a comment on that page with the same error. Actually, there are quite a few errors listed in the comments.

I'm unfamiliar with the toolbox and we don't have much info here. If the demo is just meant to be run without any user inputs then the error may be due to an incompatible matlab release/version.

Answer 1

[Chirdpong Deelertpaiboon]

I hope this is not too late for your question. I just found the answer from this post

https://www.mathworks.com/matlabcentral/answers/405270-error-using-classreg-learning-fullclassificationregressionmodel-preparedatacr-line-192-x-must-be-a

This works for me. There are two lines from the example code that need to correct. The first one is

trainingFeatures = activations(convnet, trainingSet, featureLayer,'OutputAs','rows');

The second one is

testFeatures = activations(convnet, testSet, featureLayer,'OutputAs','rows');

Answer 2

[Melvyn Smith]

Yes, I see those comments also. That page does not seem to be a forum for getting help (lots of questions but no answers). Actually, there does not seem to be much help and guidance available generally for MATLAB and deep learning. I've only found a couple of books and nothing really good on-line. Maybe MATLB is not the way to go and something like Python or Java is better. I am trying to get into CNNs. I read several books and feel I know the underlying concepts. Unfortunately, the valley of death seems to be going from that to coding CNNs for one's own applications.