plsmod: metrics for multiple outcomes?

[Jannik]

Hi, I have successfully built a pls model in the tidymodels framework using the plsmod package.

I would like to tune the pls parameters "predictor_prop" and "num_comp" (and some recipe stuff) using resamples, but metrics for multiple outcomes is not yet supported by tidmodels I understand...?

Has anybody written a custom metric for this use case that I can copy?
Or is there already a package that does this?

Thanks

/Jannik

[Max]

We don't have things setup for multivariate analysis yet. parsnip can fit them but no other parts of the infrastructure can (and it is not at the top of the development list right now).

[Jannik]

Thanks Max, I understand.

Is there a way I can create my own metric for tuning purposes?
I.e. Documentation showing how to, and what to consider?
(I'm modelling growth curves, I.e. All 300+outcomes are on the same scale, so I was thinking I could just return one number (rmse) per sample)

Thank you for developing this exciting framework

/Jannik

[Max]

@Jannik

Is there a way I can create my own metric for tuning purposes?
I.e. Documentation showing how to, and what to consider?

Yes, there is documentation here.

[Jannik]

Thanks Max, I will give it a try!