Packages

class MDP extends AnyRef

The MDP class provides Markov Decision Processes.

Linear Supertypes
AnyRef, Any
Ordering
  1. Alphabetic
  2. By Inheritance
Inherited
  1. MDP
  2. AnyRef
  3. Any
  1. Hide All
  2. Show All
Visibility
  1. Public
  2. All

Instance Constructors

  1. new MDP(t: Tensor3D, r: Tensor3D, g: Double = 0.9, thr: Double = 1E-4)

    t

    the transitional conditional probability t(s1, a, s2) = P(s2 | s1, a), where s1, s2 in S, a in A

    r

    the reward tensor r(s1, a, s2)

    g

    the discount factor in (0, 1)

    thr

    the threshold (stop when v changes little)

Value Members

  1. def getPi: VectoI

    Return the optimal policy, i.e., the best action for each state.

  2. def optimize: (VectoD, MatriD)

    Determine optimal values for the state value function 'v' and the state-action value function 'q'.