Markovian Decision Rule

A Markovian Decision Rule is a Decision Rule that determines what action to take based on former decisions history for all possible states.

• AKA: Randomized Decision Rule.
• Example(s):
  • …
• Counter-Example(s):
  • a Hierarchical Decision Rule,
  • a Minimax Decision Rule.
• See: Markov Decision Process, Decision Epoch, Randomized Experiment.

---

---

References

2017

• (Sammut & Webb, 2017) ⇒ Claude Sammut, and Geoffrey I. Webb. (2017). "Markovian Decision Rule”. In: (Sammut & Webb, 2017).
  • QUOTE: In a Markov decision process, a decision rule, [math]\displaystyle{ d_t }[/math], determines what action to take, based on the history to date at a given decision epoch and for any possible state. It is deterministic if it selects a single member of [math]\displaystyle{ A(s) }[/math] with probability 1 for each [math]\displaystyle{ s \in S }[/math] and for a given [math]\displaystyle{ h_t }[/math], and it is randomized if it selects a member of [math]\displaystyle{ A(s) }[/math] at random with probability [math]\displaystyle{ q_{d_t(h_t)} (a) }[/math]

---