MachineLearning/mdp library

Markov Decision Process (MDP) utilities - generic

A compact, engineering-oriented MDP helper that provides:

  • generic value iteration and policy iteration solvers for discrete MDPs
  • support for arbitrary state/action indexes with transition and reward represented as dense arrays (List)
  • convergence criteria and iteration limits

Contract:

  • Input: number of states, number of actions, transition Pas' (prob), reward Ras', discount gamma.
  • Output: value function and greedy policy.
  • Errors: throws ArgumentError on shape mismatches.

Classes

MDP