Modeling Low-risk Actions from Multivariate Time Series Data Using Distributional Reinforcement Learning.