r/deeplearning - How do I do reinforcement learning in this case?

At each time step, I need to give a score to a varying number of items n. Depending on the score some different things will happen to the i best and j worst items. I need to optimize the neural network that does the scoring through time.