The clean version of alibi-test.
Only contains two-sample KS test and ChiSquare test to detect distribution drift of two samples.
This version suppurts Tensorflow and Pytorch which allows the model to run on GPUs.
from drift_detection import drift_detection
ref_logist = my_model(X_ref)
drift_detector = drift_detection(ref_logist, 0.05, method="KSDrift")
or
drift_detector = drift_detection(ref_logist, 0.05, method="ChiSquareDrift")
you can update any parameters and data by
drift_detector.update_ref(new_x_ref)
drift_detector.update_threshold(new_threshold)
get the result by
x_logist = my_model(x)
preds = drift_detector.get_result(x_logist)