metrics

[gzt5142]

Currently, HyTEST is coding statistical metrics as functions in notebooks, and these notebooks are called into other notebooks for use in various workflows. Using metrics in notebooks is a useful way to 'import' common functions so that we don't duplicate code all over the place. Sourcing those common functions using a %run magic seems to work for us in a tutorial context, and it lets us put good descriptions (in markdown) along side the code cells (including equations and references/citations). And this publishes well in the jupyterbook (IMO).

At some point, you may want to consider building a stand-alone python package for metrics that you want to use outside of the tutorial context. We talked about this a loooooooonnnnng time ago ("pydroGOF" as a python replacement/port of the hydroGOF R package's functionality).

I think it's worth considering as we put more and more common functions into shared space to economize on code re-use. See the scikit-learn metrics sub-package as an example of how the ML folks are doing something similar. If it is well-designed, the hytest.metrics module would give a hydro-specific flair to what is otherwise generic computations.

This would be a good fit for the "domain-specific" layer in Jim Bednar's SOSA model: https://github.com/panstacks/pandata

We want to find the best way forward, whether that is contributing to another open-source python hydrological stats package or something like suggested above.

[jamaa]

You might also want to consider using or building on an existing package such as hydroeval or hydroerr

[sfoks]

I like this idea of adding back to a published package, it looks like the last commits to these packages were several + years ago. We may need to shop around to find the right fit but I think your suggestion is great. Thanks for posting!

[ThomasMGeo]

If you can adhere to the scikit-learn interface where possible, and maybe use units via Pint would be great!