Metrics for the designed network

[barneydobson]

Description

Add a set of new metrics and tests, all reasonably simple ones.

Fixes #91

This does highlight a possible issue that I am likely re-evaluate dominant_outlet and best_outlet_match many many times during iterate_metrics. I don't think this would be constraining, but it does feel wasteful - I have created an issue to address it #101.

[dalonsoa]

LGTM.

Yeah, it seems a bit wasteful. If the inputs were hashable, you could cache the result, but I don't think that's the case, so we would need to come up with a more creative way of avoiding calculating the same thing again and again.

[cheginit]

Since bias has a sign, for the three pbias functions please make a note in the docstring that this is sim - real or even better, provide the formula using a math block it shows up nicely in RTD.

[cheginit]

My only concern with the metrics is handling NaNs. Do you any checks in the pre-processing steps for removing NaNs? If not, you need to either drop them explicitly or use numpy's (or numbagg) NaN specific functions, such as np.nanmean, throughout the code base where NaNs might occur.

[barneydobson]

handling NaNs. Do you any checks in the pre-processing steps for removing NaNs? If not, you need to either drop them explicitly

I do in the NSE/KGE calcs (in the relevant branch). And I've not encountered them anywhere else (neither SWMM simulations or data files). I will make an issue to do a more thorough check though