Policy gradient ascent

The basic intuition behind PG methods is we move from finding a value function that describes a deterministic policy to a stochastic policy with parameters used to define a policy distribution. Thinking this way, we can now assume that our policy function needs to be defined so that our policy, π, can be set by adjusting parameters θ so that we understand the probability of taking a given action in a state. Mathematically, we can simply define this like so:

You should consider the mathematics we cover in this chapter the minimum you need to understand the code. If you are indeed serious about developing your own extensions ...

Get Hands-On Reinforcement Learning for Games now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.