mrkwjc/einsumt

Multithreaded version of numpy.einsum

einsumt's Issues

`einsumt` fails if the shape of (ellipsis) trailing axes are not equal

...however, np.einsum does automatic broadcasting. Any ideas on implementing that?

Minimal working example:

import numpy as np
from einsumt import einsumt

a = np.ones((3, 3, 1, 1))
b = np.ones((3, 3, 8, 100))

# this works
np.einsum("ij...,ij...->...", a, b)

# this fails
einsumt("ij...,ij...->...", a, b)

# ...this also fails
einsumt("ijmn,ijmn->mn", a, b)

Thanks for providing einsumt, it is a fantastic package. A little side-node: Mostly, einsumt is equal in execution times compared to just-in-time compiling the code (Numba). I'm using einsumt in my finite elements package FElupe.

Best regards,
Andreas