byu-hydroinformatics / hydroerr

Goodness of Fit metrics for use in comparison studies, specifically in the field of hydrology.

Goodness of fit metrics for use in comparison studies, specifically for use in the field of hydrology

• Documentation: https://hydroerr.readthedocs.io/en/stable/
• Source: https://github.com/BYU-Hydroinformatics/HydroErr
• Bug reports: https://github.com/BYU-Hydroinformatics/HydroErr/issues

HydroErr provides:

• A library of over 70 error metrics to measure hydrologic skill
• Treatment of NaN, Inf, negative, and zero values.
• Scholarly journal sources and code examples for metric methods in the documentation.

Publications:

• Roberts, W., Williams, G., Jackson, E., Nelson, E., Ames, D., 2018. Hydrostats: A Python Package for Characterizing Errors between Observed and Predicted Time Series. Hydrology 5(4) 66, doi:10.3390/hydrology5040066

We would request that if you use this package that you please cite the above paper. The journal paper is open access and can be found at https://doi.org/10.3390/hydrology5040066. BibTeX, EndNote, and RIS files can be downloaded at the journal site.

hydroerr's Issues

HydroErr R_squared vs HydroErr NSE vs Sklearn R_squared

I tried comparing the output of HydroErr's r_squared and nse with SKLearn's r2 score.

I found out that HydroErr's NSE and Sklearn's r2 scores are the same. Meanwhile HydroErr's R_squared is different from SKLearn's R2 score. Is this a bug? If not, can someone help me understand the reason for the discrepancy?

Below is a sample code:

from sklearn.metrics import r2_score
import HydroErr as he

y_true = [3, -0.5, 2, 7]
y_pred = [2.5, 0.0, 2, 8]
sklearn_r2 = r2_score(y_true=y_true, y_pred=y_pred)
hydroer_nse = he.nse(simulated_array=y_pred, observed_array=y_true)
hydroer_r2 = he.r_squared(simulated_array=y_pred, observed_array=y_true)

print("SKLearn R2: ", sklearn_r2)
print("HydroErr NSE: ", hydroer_nse)
print("HydroErr R2: ", hydroer_r2)

# # Output ##
# SKLearn R2:  0.9486081370449679
# HydroErr NSE:  0.9486081370449679
# HydroErr R2:  0.9699681653424412

As you can see above, SKLearn's R2 and HydroErr's NSE are the exact same.