Nystrom SVD Bug: eigvals argument in scipy.linalg.eigh has been deprecated

[zjf4]

Hi all,

I tried to run EigenPro 3.0 but ran into an error in the Nystrom SVD step. The error was TypeError: eigh() got an unexpected keyword argument 'eigvals'. The latest version of Scipy has deprecated this argument. The fix is easy. You just need to replace the keyword argument with subset_by_index.

vals, vecs = linalg.eigh(scaled_kmat, subset_by_index=(n_sample - top_q, n_sample - 1)).

I wanted to bring this to your attention so you can patch it.

[parthe]

@zjf4 Thanks for the correction.
Would you be able to create a pull request? I can merge it.

[parthe]

Meanwhile, do try out EigenPro 4.0 which is much faster

[zjf4]

Done!

Thank you for telling me about EigenPro 4.0. I was looking at the code there, and the same issue occurs: https://github.com/EigenPro/EigenPro/blob/main/eigenpro/utils/eigh.py

I'm happy to issue another pull request to fix it there as well.