Support for Complex-Number Tensors in PhySO: Seeking Guidance or Workaround

[LostArkRaider]

Hello PhySO Community,

Thank you for creating and maintaining this incredible tool! I'm excited to dive into PhySO, but I’ve encountered a roadblock that I believe might resonate with others working on similar use cases.

I’m currently working on a project that involves processing complex-number tensors (dtype: torch.complex64). However, I ran into an issue in dataset.py at the following assertion:

assert X.dtype == torch.float64 or X.dtype == torch.float32, "X must contain floats."

This check prevents me from directly using complex-number tensors as inputs. My use case heavily relies on complex-number tensor math, and I’m wondering if there’s an established workaround or alternate approach within PhySO to handle this type of data.

It appears that several of the Feynman Lectures on Physics formulas used for benchmarking potentially have a complex-number result. It would be helpful to know how that is handed in PhySO.

I’d greatly appreciate any guidance or suggestions from the community. This functionality would open up exciting possibilities for applications in my field, and I’m sure others exploring complex tensor use cases might benefit as well.

Thank you for taking the time to read this, and for any help you can provide!

Best regards,
Keith Brickey