FEERCI is an opinionated, easy-to-use package for calculating EERs and non-parametric confidence intervals efficiently. It offers a single method, feerci.feerci, that returns both an EER and CI for provided impostor and genuine scores. The only dependency is numpy.
pip install feerci
- Feerci should now support python <= 3.10, rather than python <= 3.6
- feer and feerci now include optional flag to return score threshold@eer.
- Fixed bug in feerci that caused segfault when impostor and genuine score lists did not have any overlap
- Switched output arguments around, to make more intuitive sense
- Initial release of package
FEERCI is distributed under the MIT license
0.3.0
Calculating just an EER:
import feerci import numpy as np impostors = np.random.rand(100) genuines = np.random.rand(100) eer,_,_,_ = feerci.feerci(impostors,genuines,is_sorted=False,m=-1)
Calculating an EER and 95% confidence interval (the default) on 10000 bootstrap iterations (the default):
eer,ci_lower,ci_upper,bootstrapped_eers = feerci.feerci(impostors,genuines,is_sorted=False)
Calculating an EER and 99% confidence interval on 10000 bootstrap iterations (the default):
eer,ci_lower,ci_upper,bootstrapped_eers = feerci.feerci(impostors,genuines,is_sorted=False,ci=.99)
Calculating an EER and 99% confidence interval on 1000 bootstrap iterations:
eer,ci_lower,ci_upper,bootstrapped_eers = feerci.feerci(impostors,genuines,is_sorted=False,m=1000,ci=.99)
FEERCI & FEER can both calculate the threshold @ the EER:
eer,ci_lower,ci_upper,bootstrapped_eers = feerci.feerci(impostors,genuines,is_sorted=False, return_threshold=True)