Analysing Factorizations of Action-Value Networks for Cooperative Multi-Agent Reinforcement Learning