SacActorHead
Enums · AiDotNet.Enums
Selects how a Soft Actor-Critic policy produces the SPREAD of its action distribution — the standard deviation of the Gaussian it samples from.
For Beginners
The mean is the position the agent wants to take; the spread is how much it is willing to jitter around it while exploring. The question this setting answers is whether that jitter is one fixed amount for the whole run, or something the network decides per state.
How It Works
SAC’s policy is a diagonal Gaussian, pi(.|s) = N(mu(s), diag(sigma^2)). Both members below share
the same mean head mu(s); they differ only in where sigma comes from, which is what decides
whether the agent can be MORE uncertain in some market states than others.
Fields
| Field | Summary |
|---|---|
SacActorHead StateConditionedGaussian |
The actor predicts the log standard deviation from the state, alongside the mean. |
SacActorHead StateIndependentLogStd |
One learned log standard deviation per action dimension, shared across every state. |
int value__ |