a) Imitation learning is more likely to be effective (compared to other forms of RL) when the transition matrix and/or the rewards are stochastic