Markov decision processes#

Protocol#

`agent.MDPAgent`	Base class for agents in the discrete MDP setting.
`environment.DiscreteMDP`	Finite Markov decision process with no exploitable structure.
`interaction.MDPInteraction`	Interaction loop for the discrete MDP setting.

`agents.IMED_RL.IMEDRL`	Indexed Minimum Empirical Divergence Reinforcement Learning (IMED-RL).
`agents.PSRL.PSRL`	Posterior Sampling Reinforcement Learning.
`agents._Oracle.build_opti`	Build the oracle for an environment.
`agents._Oracle.Opti_controller`	Oracle that solves the MDP exactly and follows its optimal policy.
`agents._Oracle.Opti_swimmer`	Hand-coded oracle for RiverSwim: always swim right.
`agents._Random.Random`	Uniform exploration: take a random action in every state.
`agents.Human.Human`	Interactive agent that asks the user for each action.
`agents.Human.keyboard_waitfor`	Block on stdin until the user types one of the allowed strings.

`envs.riverswim.RiverSwim`	The RiverSwim hard-exploration benchmark MDP.
`envs.riverswim.ErgodicRiverSwim`	RiverSwim variant in which swimming left may still drift right.
`envs.randomMDP.RandomMDP`	Randomly generated finite MDP with sparse transitions and rewards.

Print each MDP step to stdout as a colourized row of states.