In this dissertation, we aim to adapt reinforcement learning (RL) to real-world, high-risk settings. We study how to optimize sequential decision-making in complex settings with large observational data repositories where ...