In this work, we evaluate machine learning (offline) and evolutionary strategy (online) techniques for the Raman pump power optimization. Experimental results show that, although reusable and accurate, online tools may be time-consuming for reconfigurable amplifiers.