Return to Answer

deleted 1 character in body

Source Link

edited Dec 12, 2020 at 12:05

41.4k
12
114
205

As long as your policy (propensity) is differentiable, everything's is good. Discrete, continuous, other, doesn't matter! :)

A common example for continuous spaces is the [reparameterization trick][1]reparameterization trick, where your policy outputs $\mu, \sigma = \pi(s)$ and the action is $a \sim \mathcal{N}(\mu, \sigma)$.

As long as your policy (propensity) is differentiable, everything's is good. Discrete, continuous, other, doesn't matter! :)

A common example for continuous spaces is the [reparameterization trick][1] where your policy outputs $\mu, \sigma = \pi(s)$ and the action is $a \sim \mathcal{N}(\mu, \sigma)$.

As long as your policy (propensity) is differentiable, everything's is good. Discrete, continuous, other, doesn't matter! :)

A common example for continuous spaces is the reparameterization trick, where your policy outputs $\mu, \sigma = \pi(s)$ and the action is $a \sim \mathcal{N}(\mu, \sigma)$.

Source Link

created Dec 12, 2020 at 6:22

kaiwenw

As long as your policy (propensity) is differentiable, everything's is good. Discrete, continuous, other, doesn't matter! :)

A common example for continuous spaces is the [reparameterization trick][1] where your policy outputs $\mu, \sigma = \pi(s)$ and the action is $a \sim \mathcal{N}(\mu, \sigma)$.