Use numba to accelerate OpticalThickness calculation? about pyrads

[brian-rose]

I haven't thought much about it beyond musing that acceleration with numba is probably a good idea :-)

It would be great to make some progress on #6 and add some tests to ensure that the package is portable and reproducible on different platforms. Acceleration without tests could be a recipe for trouble.

[brian-rose]

The fortran wrapping can be a chore. The RRTMG wrapper for climlab could serve as a useful example.

[brian-rose]

@AndrewWilliams3142 it's not pretty! If I can find some time I will take another look at #6. I do think the automation of the build is important to make this package more portable and more testable.

[AndrewILWilliams]

Good point @brian-rose ! I'll start working on some tests this week.

I've also added a setup.py in #13 , which seems to work when I tried it locally. I'm not sure how to wrap fortran build though.

[danielkoll]

Numba does seem to speed things up significantly -- great work @AndrewWilliams3142 !

I've added a conda env so new PyRADS users can hopefully use this out-of-the-box. @brian-rose @AndrewWilliams3142 let me know if you have any other suggestions to increase user friendliness

[AndrewILWilliams]

No worries @ddbkoll , happy to help!

As @brian-rose said already, I think the next step should be to automate the fortran build ( and fully close #6 ), this will make it easier to add some unit tests as well. I did a dive into the climlab RRTMG wrapper as suggested, but I don't really know enough about conda/compilers to be able to follow the details sadly

[AndrewILWilliams]

@brian-rose great! I just noticed your minimalf2py repo actually - I'll take a look at that and see if I can understand the automatic compilation process a bit better. Happy to be of help if i can

[brian-rose]

Yes, that's a good place to start (and I should have mentioned earlier). I created that repo specifically to help debug some build / CI issues.