Look at M5 partitioning algorithm

[benelot]

learns the partitioning of the state space in reinforcement learning

[benelot]

Also look at the algorithm that was used in the Deepmind paper to flatten the space to 2D.