Comparing the Efficacy of Fine-Tuning and Meta-Learning for Few-Shot Policy Imitation