MPS implementation

[khurramscience]

No description provided.

[khurramscience]

MPS implementation showed great improvements

[wouterwln]

Hey @khurramscience, thanks for the interest in the paper and for the deep dive on the big tensor scaling issue.

The Tensor Train/MPS approach is a solid idea (and one I was unaware of). Using the Rényi entropy trick to bypass the Shannon summation is a clever move for this kind of state/observation space.

A few quick things on the PR:

• This repo is a specific artifact for the paper's experiments, so I want to keep the core code as-is to ensure reproducibility for people reading the original work, and to keep the code to implement the exact idea from the paper.
• The PR now adds a .zip file, I can't really review or merge a .zip file for security and version control reasons.
• If you want to put the code in a separate public repo (or want me to host one on my personal/biaslab github), send me the link. I’d be happy to check it out there (and maybe I can contribute), and if that approach helps scale the next phase of this project, I'd be more than happy to credit you for the idea/implementation.

Cheers,
Wouter