Epsilon soft policy

---

tags:

• 🌱
• AI
• ComputerScience date: 20--Feb--2023

---

Epsilon soft policy

Related to Soft policy

Different from the greedy approach where each action has at least $p = \frac{\epsilon}{|A|}$ chance of being selected. $$\pi(s) =
\begin{cases} p=1-\epsilon+\frac{\epsilon}{|A|}, & \alpha=\alpha^* \ p=\frac{\epsilon}{|A|}, & \alpha\ne\alpha^* \end{cases}$$

---

Links: