Policy Evaluation

The evaluate function provides a policy evaluation tool for MDPs:

POMDPModelTools.evaluate — Function

evaluate(m::MDP, p::Policy)
evaluate(m::MDP, p::Policy; rewardfunction=POMDPs.reward)

Calculate the value for a policy on an MDP using the approach in equation 4.2.2 of Kochenderfer, Decision Making Under Uncertainty, 2015.

Returns a DiscreteValueFunction, which maps states to values.

Example

using POMDPModelTools, POMDPPolicies, POMDPModels
m = SimpleGridWorld()
u = evaluate(m, FunctionPolicy(x->:left))
u([1,1]) # value of always moving left starting at state [1,1]

source