Initial state priors#
The initial_state_prior entry of a state factor gives its distribution at the start of an
episode. It accepts:
over discrete levels:
Categorical, orDirichletwhen the prior is learned, and their conditional formsConditionalCategoricalandConditionalDirichletwhen the prior has parents;over a continuous state:
GaussianorMixtureGaussian;a learned starting value for a carried deterministic node:
LearnedVector;the state factor’s own transition evaluated at the start:
DerivedInitial.
Declarations specific to priors#
"This state factor's initial belief is its own transition, at the episode start." |
|
A trainable vector behind an activation: a parametrized constant, with no parents. |