Re: Allow matrix multiplication in structured arrays

Hi,

Thanks for replying. I was using:
```
from numpy.lib import recfunctions as rfn
xyz = rfn.structured_to_unstructured(atoms[["x", "y", "z"]])
xyz = xyz @ transform.T
atoms[["x", "y", "z"]] = rfn.unstructured_to_structured(
    xyz, dtype=atoms[["x", "y", "z"]].dtype
)
```
But I think this is creating a copy not a view.

Anyway, I did the following:
```
fields = atoms.dtype.names
idxx, idxz = fields.index("x"), fields.index("z")
xyz_field = ("xyz", (np.float64, (3,)))
# create a new dtype with previous fields, xyz and next fields
new_fields = (
    [(n, atoms.dtype[n]) for n in fields[:idxx]]
    + [xyz_field]
    + [(n, atoms.dtype[n]) for n in fields[idxz + 1 :]]
)  # usually x, y, z fields are one next to the other
new_dtype = np.dtype(new_fields)

xyz = atoms.view(new_dtype)["xyz"]
xyz[:] = xyz @ transform.T
```
And works fine. I guess there is no direct way computing the rotation without a view.

Thanks Marten,

Best,
Abel
_______________________________________________
NumPy-Discussion mailing list -- [email protected]
To unsubscribe send an email to [email protected]
https://mail.python.org/mailman3//lists/numpy-discussion.python.org
Member address: [email protected]