`lt` does not support torch.complex Tensors

[Michael Poli]

To reproduce

import torch
import lovely_tensors as lt
lt.monkey_patch()

x = torch.randn(4, dtype=torch.complex32)
x

Gives

RuntimeError: isposinf does not support complex inputs.

[Alexey Zaytsev]

Hi. Thank you for the report.

Do you have a concrete use case for complex tensors?

I will make sure lt can handle them, but do you have ideas on what a concise and informative representation of a complex tensor should look like?
Besides a text summary, is there a graphical representation that would make sense?

[Alexey Zaytsev]

I've pushed a quick fix that just displays the plain version for complex tensors.

# We don't really supposed complex numbers yet
c = torch.randn(10, dtype=torch.complex64)
c

tensor([ 0.5135+0.0645j, -0.2751+0.3733j, -0.2552-0.0428j,  0.0518+0.5789j,  1.0469+0.2439j, -1.0070-0.0823j,
     0.1538-0.0330j, -1.0137-0.4006j, -0.3007+0.1856j, -1.0176+0.3687j])

Let's keep the issue open because there is definitely room for improvement.

[Michael Poli]

A simple solution would be to treat real and imaginary parts of the tensor as two different tensors, perhaps using torch.view_as_real

[Alexey Zaytsev]

A simple solution would be to treat real and imaginary parts of the tensor as two different tensors, perhaps using torch.view_as_real

Could you suggest a codebase, ideally a notebook, that is heavily using complex tensors? I feel that I still need to better understand how people use them in real code.