Distributions#
Beliefs, priors and parameter posteriors are distribution objects: immutable equinox modules
that hold their parameters as arrays. The conditional families give a child distribution for every
configuration of its parents, either by table or by a small network. Exponential-family members
expose their natural and mean parameters through one shared contract.
Base types#
Abstract base for every belief, message payload and preference object. |
|
A |
|
Structural capability protocol (see module docstring), satisfied by method presence. |
|
Abstract base for conditional families: Markov kernels P(child | parents) with discrete parents. |
Distributions over states#
A categorical distribution over |
|
Joint distribution over |
|
A multivariate Gaussian with full covariance, batched over leading axes. |
|
Gaussian mixture belief. |
|
A diagonal Gaussian |
|
A Dirichlet distribution over the |
Conditional families#
Tabular CPD P(child | parents): one simplex per parent configuration. |
|
Parent-indexed stack of Gaussians: P(x | parents) with discrete parents. |
|
Independent Dirichlet priors over a CPD, one per parent configuration. |
|
MLP over concatenated parents → a factored categorical child. |
|
MLP over concatenated parents → diagonal-Gaussian child (see module docstring). |
|
MLP over concatenated parents → a raw output vector, with no density of its own. |
|
An |
|
A trainable vector behind an activation: a parametrized constant, with no parents. |
Parameter priors and further families#
These are exported by hybrid_aif.distributions only.
|
|
Independent inverse-Gamma posteriors over per-dimension noise variances. |
|
Inverse-Wishart posterior over a full noise covariance. |
|
Gaussian mixture whose component covariances are diagonal plus low rank, |
|
A GRU carry update |