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.
First use model to predict the label
ref_logist = my_model(X_ref)
x_logist = my_model(x_h0)
Then call the function drift_detection, and select "KSDrift" or "ChiSquareDrift" in the method
preds = drift_detection(ref_logist, x_logist, 0.01, method="KSDrift")