Ilhan Polat

This error comes from ARPACK and the authors in the error message is actually ARPACK authors. The error often is generated by the routine `dneupc`, `dneupd` etc. It usually stems...

@MrYann if you don't mind taking a stride in the memory lane, we are finally addressing these in #22748 . I was wondering if you still have these matrices lying...

Thank you for checking @MrYann anyways. I think we addressed this now in the rewrite which I was getting once in a while.

Matt if you have any capacity at all, could you try the hanging test runs again over #22748 ? I tested on different machines via pip but my conda build...

For my taste this is indeed neat ...But: In my opinion the added value is limited since nobody chooses the method on the fly that often. Remembering which method to...

I think we should also consider the testing mechanisms we have. There is no transparency in what is being tested and with what data points.

> I had expected inplace updates to be more performant. They are. In this case b[mask] is doing an advanced indexing hence returning a copy then assignment is also slow...

Here is the ballpark of performance when advanced index removed ```python N = 10_000_000 # 80 MB P = 0.05 import numpy as np a = np.random.random(N) b = a...

True but this functionality requires a true in-place operation which is needed in many parts of the codebase and maybe NumPy should reconsider bool mask handling in a different fashion....

Is this failure also happening on Ubuntu box? I've tested the problematic case in the SciPy issue on Win10, Win11, Fedora and Mint and could not reproduce. I could only...