Closed leocd91 closed 2 years ago
This is what we do in the current version DeepAL. This greedy approach usually works in practice. Of course there must be some better ways to handle the batch scoring but currently we do not focus on it.
ah, i see, I get it after seeing your code. thanks.
hi @ej0cl6 , why KCenterGreedy use training data to get the embeddings
? why there's no unlabelled data being included there?
It is used for learning the centers.
Suppose it's a binary problem, did I do it right if I just sort it directly the prob or uncertainties result?