Exploratory Gradient Boosting for Reinforcement Learning in Complex Domains