How to integrate customized kernel function into "regression learner" toolbox?

The regression learner in 2017 contains SVM method with linear, rbf, and poly kernels. What if I want to use my own kernel? Is there a way that I can integrate my own kernel to the toolbox so that I can use it easily?

Answer 1

[Mukul Rao]

Hello,

The "Regression Learner" app does not currently support specifying custom Kernel functions. However, if you are willing to consider a command-line approach, you could use the fitrsvm function to fit a Regression Support Vector Machine. In the inputs for fitrsvm, you can specify the "KernelFunction" Name-Value pair to point to your custom Kernel function. Please refer the following link for more information:

https://www.mathworks.com/help/stats/fitrsvm.html#namevaluepairarguments

As an alternate workaround, you can generate code for your regression model from the Regression Learner App using the "Export Model" drop down. You can modify the "KernelFunction" property in the generated code to point to your custom Kernel function. You might have to also modify other Kernel associated parameters such as "KernelScale" accordingly as required by your use case.

On a different note, I work for the MathWorks and I have forwarded this use case to the appropriate product team.

[xi]

Thanks for your answer and thanks for forwarding this to the product team. I know how to apply customized kernels with command-line codes, but I'm more inclined to take advantage of the "regression learner app", which is very convenient to use. Is that app open source?

Actually, I'm interested in designing an app similar to the regression learner, but also includes the feature mentioned in my question. I never used matlab app designer before. Just wondering if there's short-cut I can go without going through all the development process?

[zhaorong huang]

It's already 2023. Is there a more detailed instance of the calling code available for use

Answer 2

[Chibuzo]

This has been implemented in the 2023b version (or earlier):

https://se.mathworks.com/help/stats/support-vector-machines-for-binary-classification.html#buax656-1