Most of the time, you don’t need to re-learn. Once smart:comp 3 has analyzed your signal, the learned information stays valid. For example, switching profiles does not require re-learning.The learned observation of your signal remains unchanged - the profile simply interprets that information differently.
You only need to re-learn when the audio itself changes noticeably (different performance, different take), or if you feel that smart:comp 3 analyzed a non-representative part of the signal and you want it to “listen” to a better section.