Andrew Knyazev, Professor Emeritus
Andrew Knyazev, Professor Emeritus
Using `matmat` substituting `matvec` for dtype determination fails above; see https://github.com/scipy/scipy/issues/19207#issuecomment-1712842372 So revert the change.
I have also tried to include `rmatvec` or `rmatmat` in dtype determination in ``` #if self.rmatvec is not None: # v = np.zeros((self.shape[0],), dtype=np.int8) # rdtype = np.asarray(self.rmatvec(v)).dtype # self.dtype...
Ready for review please
@rkern I guess that I should also update in scipy/sparse/linalg/_eigen/arpack/arpack.py the following ``` def _aslinearoperator_with_dtype(m): m = aslinearoperator(m) if not hasattr(m, 'dtype'): x = np.zeros(m.shape[1]) m.dtype = (m * x).dtype...
> I don't think so. `LinearOperator` objects will always have a `.dtype` attribute set (unless if one writes a particularly broken subclass). Maybe remove it then as never being executed?...
> Oh, that's the code that already exists? Yes, it should be removed and replaced with just `aslinearoperator()` wherever it is called. It's a very old function that predates the...
Could someone please review it? It's completed.
This PR is 4 months old now. Could someone review it and consider merging?
I already forgot about this PR... Could someone please review it, otherwise it might just slip through the cracks again.
It is back to life and ready for the final review and merge please.