Inflow probability vs vehs_per_hour

[kjang96]

When adding new Inflows with the probability param=1/12 veh/s, as opposed to vehs_per_hour=300 (note that probability=1/12 will result in roughly 3600s/12veh=300 vehicles per hour), training rllab/green_wave.py appears to learn nothing at all for ~30 iterations. This is compared to vehs_per_hour which learns almost linearly (albeit slowly) in the same amount of time. I'm wondering if this is an Inflows problem on our end or with Sumo