Deep Reinforcement Learning for the Co-Optimization of Vehicular Flow Direction Design and Signal Control Policy for a Road Network.