The package contains a finite horizon MDP solver for discrete problems. This algorithm is a modified version of infinite horizon value iteration.
You must have POMDPs.jl installed. To install
FiniteHorizonValueIteration, run the following command:
using POMDPs using Pkg POMDPs.add_registry() # TODO: Is this still needed? Pkg.add("FiniteHorizonValueIteration")
using FiniteHorizonValueIteration mdp = MyMDP() # initialize MDP # initialize the solver solver = FiniteHorizonSolver(verbose=false) # run the solver policy = solve(solver, mdp)
Output and validation
The policy is stored to
Example problems are defined in
test/instances/..., examples are used in corresponding test files. Results are validated against the value iteration executed on all epochs simultaneously.