Change goals for RUNE, python framework with Rust backend?...

[wilhelmagren]

I think restructuring the repo to feature a high-level Python interface over the Rust backend is more attractive than simply creating another Rust tensor crate...

The underlying backend functionality would still be usable by itself, and concepts could be easily transferred to other projects.

What we are interested in speeding up is:

• Tensor operations
• Backpropagation
• Data transformations

And maybe most interesting is to minimize number of dependencies, keeping RUNE purely minimalistic. The goal is to be able to use it on microchips with extremely limited disk space, and be efficient enough to perform real-time interference.

Some (expected) required dependencies:

• numpy (Python) [link]
• numpy (Rust) [link]
• ndarray (Rust) [link]

Other than that, most likely networkx or even rustworkx for DAG visualization...