Hi, thanks a lot for the amazing library!

Currently, tensorly.contrib.sparse.decomposition provides a robust power method for symmetric sparse tensors, which is great. However, it seems there is no support for the general non-symmetric case (e.g. sparse tensors of shape k × k × d that are not fully symmetric).

Would it be possible to add a version of parafac_power_iteration that supports sparse tensors without assuming full symmetry?

This would be very useful for applications like: Hidden Markov Models (minimal realization).

Right now, the only workaround is to convert to dense, which doesn’t scale well.

Thanks again for your great work!