/MDPSolver

Primary LanguageC++

MDP Solver

MDP solving program that uses a variety of different algortihms. Currently supports:

  • Value Iteration (VI)
  • Policy Iteration (PI)
  • Modified Policy Iteration (MPI)
  • Linear Program Formulation & Solution (using gurobi) (LP)
  • TD(0) Policy Evaluation (TD0)
  • Every-Visit Monte-Carlo Policy Evaluation (EVMC)

Command Line arguments:

  • -f, -file: The file to read from (default = 'mdp.csv')
  • -m, -method: The method to use. Choose from the above abbreviations (default = 'VI')
  • -g, -gamma: The discount value (default = 0.9)
  • -a, -alpha: The alpha value to use (e.g for EVMC) (default = 0.1)
  • -i, -iterations: The number of iterations for the algorithm (default = 1)
  • -r, -runs: The number of runs (total = runs*iterations) (default = 1)
  • -t, -threshold: The threshold for VI, PI etc. (default = 0.0001)