Weighted PLS

[nic-Oban]

Hello,
thanks for this fine tuned pls that is very fast!
I searched for but can't find anything on weighted pls. Is there any optimized option to apply a diagonal matrix of sample weights in your algorithm? (as in Schaal et al. 2002)
In WPLSR, a priori weights, different from the usual 1/n (standard PLSR), are given to the n training observations. These weights are used for calculating (i) the scores and loadings of the WPLS and (ii) the regression model that fits (by weighted least squares) the Y-response(s) to the WPLS scores.
Thanks a lot and have a good day.
Nicolas

[Sm00thix]

Hi Nicolas,

I'm glad you like the software :-)

I am unsure whether you are asking for weights for the regular PLS implementations in ikpls.numpy_ikpls and ikpls.jax_ikpls_alg_1 and ikpls.jax_ikpls_alg_2 or the fast cross-validation implementation in ikpls.fast_cross_validation.numpy_ikpls.

For the regular PLS implementations, adding the option for class weights is definitely doable. I will implement this in the future. However, as I am currently busy with other aspects of my Ph. D. programme, it might be a while. In the meantime, should another developer make the required modifications to the code. I am happy to merge them given they follow the Contribution Guidelines.

For the fast cross-validation, it remains an open problem whether class weights can be supported without compromising the superior computational complexity of the fast algorithm. However, I suspect that it is actually possible to solve this problem. I will definitely try and tackle it in the not-so-distant future.

Best,
Ole

[nic-Oban]

Hello Ole,
thanks for your answer.
As a first step, I'd like to weight the numpy_ikpls but the CV-one would be a great one too!
I'll have a look at the nupy_ikpls if I find some time. Thanks a lot again and have good day.
Nicolas

[Sm00thix]

Hi Nicolas,

Sounds good :-)
I will keep this issue open until someone implements weights. I will do it myself when I have time if no one beats me to it.

Have a nice day!

Best,
Ole