[DISC]: loosen up the requirements.txt

[asreimer]

The requirements for this package seem overly strict, for example, numpy >=1.16. Are these really necessary? Perhaps we can loosen the restrictions a bit?

[ljlamarche]

This probably is excessively strict. Is there any rule of thumb for how to handle these things, or should I just see how old of a numpy version I can have before things start to break?

[asreimer]

Here's some recommendations:

• numpy >= 1.16: Remove this. Matplotlib and scipy versions determine the version of numpy needed.
• h5py >= 2.9: Looking at the release history this version looks fine. Going to older versions would only allow support for older versions of Python 3 (like 3.2), but it is unlikely anyone doing data analysis is using 3.2. Probably, we don't need to specify a version number.
• matplotlib >= 2.2.4: Looking at the release history, 2.2.4 is probably fine, but I would recommend dropping the version number
• cartopy >= 0.17: Cartopy is a relatively new project, 0.17 is probably the oldest version I would recommend. I don't have a good reason for this, but I think specifying the version number here makes sense.
• scipy >= 1.2: mangopy is only using interpolate and convex hull code from scipy. I would change the minimum version to 0.17.
• future >= 0.16: Remove this. No one should be using Python 2 anymore. Numpy, matplotlib, scipy, etc don't support Python 2

[ljlamarche]

I ended up doing this in d584b4aa13672211d2854ba70a9cc9cae1eace9a as part of getting mangopy to build on readthedocs. I adopted all of @asreimer's suggestions above except for cartopy. When installing cartopy via pip, you have to first manually install several dependancies. This isn't necessary when installing cartopy with conda. If cartopy is included in requirements.txt, it produces build errors. The README does specify that cartopy must be installed BEFORE mangopy, so I think it's fine to leave cartopy out of requirements.txt, and just make the installation instructions:

1. Install cartopy
2. Install mangopy