…ist.

…ODO: Move reading net to init routine.

…rameterisation via FTorch.

* Implements reshaping of the data to the form expected by the net as a starting point.
  * Consider using torch layout argument to avoid transposition at a
    later date.

* Does not implement any normalisation of features as we need to discuss
  how to do this on a chunked grid when data is provided for the global
  grid.

* Not currently called from anywhere.

…variables from file.

…nd finalize routines.

…sing pure python (Same result obtained).]

…er than running the physics scheme beforehand.