|
If you use kNN as a learner, do the training set and testing set need to have the same class distribution? I.e., for kNN it is fair to e.g. downsample your training set to obtain a balanced dataset. Does this imply that also your test set needs to be balanced? |
