Tidy repo

[ElliottKasoar]

Resolves #24

Initial work tidying repository, so far including:

• Creating separate directories) for each benchmark code, and for each model with its data
  • I kept the model directories separate to the benchmark directories, as the models may be shared between different benchmark codes, as is the case for cgdrag/MiMA, but it might be tidier to have them in the same directory?
• General tidy up of white space etc.
• Adding new assert to interface to update cgdrag's assertions

[ElliottKasoar]

@jatkinson1000, there are a couple of unused variables/functions in the MiMA codes e.g. level, error_mesg. Is it ok to remove them or do we want to keep these to keep this code as close as possible to the original implementation?

[jatkinson1000]

No, fine to remove I think.
If unused won't make a difference.
It's not actually in the original here so not sure how/why it got there...

[ElliottKasoar]

Discussed at various stages with @TomMelt, but the general logic behind the changes to real definitions, which are the most substantial changes made in this PR, is:

• In general, I've tried to use wp where sensible for data that interacts with the model. dp is used explicitly for the timings (even when wp=dp), as there's no need for them to match in general.
• It's slightly messy, but I set wp=dp for cgdrag and MiMA, to allow the saved models to run as is, while maintaining the same style of wp for variables that in principle we may want to change the precision of together.
• There is potentially room to tidy up the assert subroutines further, either by:
  • Changing the cgdrag/MiMA benchmarks to sp (or the large_stride and resnet benchmarks to dp), such that they all share the same wp. This would allow the utils asserts to all use the same wp, but would require non-trivial changes, including updating the saved model(s), data types in various places etc.
  • Implementing templating or other methods to define the subroutines for different multiple kinds (see: How to implement same procedures for different numeric kinds fortran-lang/stdlib#35, which references an example of the use of fypp)

[ElliottKasoar]

Is the comment I added enough to close #2?

[TomMelt]

@ElliottKasoar fypp looks like a good solution for the wp=sp,dp... problem. It could be implemented in a future PR.

[TomMelt]

Thanks @ElliottKasoar. These are some really good improvements.

[TomMelt]

Is the comment I added enough to close #2?

Hard to say. I am not 100% sure. But I think it would refer to people who maybe use their distro shipped version of python. On Ubuntu, for example, you need to install dev packages separately. Something like sudo apt install python-dev. @SimonClifford is this what you were referring too?

[ElliottKasoar]

Is the comment I added enough to close #2?

Hard to say. I am not 100% sure. But I think it would refer to people who maybe use their distro shipped version of python. On Ubuntu, for example, you need to install dev packages separately. Something like sudo apt install python-dev. @SimonClifford is this what you were referring too?

I certainly encountered an error before installing python-dev on Ubuntu, which I was hoping could be inferred from the comment, but I thought it might be best to keep the comment slightly more general, as I imagine there could be similar requirements on a different distribution/OS.