Closed jarq6c closed 2 years ago
The only 'problems' I found were stylistic preferences, which is purely subjective.
I would like to eventually see quick documentation on what the metrics actually measure, instead of just directions to the papers, but that's just because I'm not a scientific user.
Thanks for the feedback! I decided to significantly expand the descriptions of all the metrics. Hopefully these new descriptions will help users unfamiliar with some of these.
This PR implements a generic
mean_error_skill_score
method and adds three new metrics:coefficient_of_persistence
,coefficient_of_extrapolation
, andvolumetric_efficiency
. These three metrics are computed in a fashion similar to the Nash-Sutcliffe Model Efficiency, except instead of using the mean baseline they use a persistence, an extrapolation, and zero baseline. A "persistence" baseline uses the observed value given a lag. An extrapolation baseline uses the last two observed values to extrapolate a linear approximation.Added @hellkite500 and @aaraney may be interested.
Additions
mean_error
mean_squared_error
root_mean_squared_error
mean_error_skill_score
coefficient_of_persistence
coefficient_of_extrapolation
volumetric_efficiency
Removals
Changes
nash_sutcliffe_efficiency
is now based on the genericmean_error_skill_score
nash_sutcliffe_efficiency
Testing
Notes
Todos
Checklist