Sample from constructed probability distribution without having to normalize

Question

Problem: I am a bit stuck on how to speed up a probabilistic algorithm. I am trying to write an algorithm that creates a random network based on Preferentia