Learning Policies for Contextual Submodular Prediction | Read Paper on Bytez