Closed annashcherbina closed 8 years ago
Thanks for the proposed change! Could you please tell me what has caused numerical differences in scores that should be equal in your case? As the scores come from a single model they are usually either identical or if not then they are different for a reason. For instance, Naive Bayes can give you scores very close to zero, such as 1e-20 and 1e-30, and these are different for a reason, and they should not be treated as ties. Therefore, I would prefer not to check equality with an epsilon-thresholded absolute difference. Thanks again for raising this important issue!
replaced != with np.close to check for equality of floats to within epsilon precision.