Parameterization
In BayesBay, all free parameters of an inference problem (see Prior) should be encapsulated within one or more instances of ParameterSpace (or, alternatively, of Discretization), which are used to define a Parameterization. ParameterSpace serves as a specialized container that not only groups an arbitrary number of free parameters but also (i) determines their dimensionality, and (ii) specifies the perturbation functions used to propose new model parameters from the current ones at each Markov chain step.
Compared to ParameterSpace, the Parameterization object is simpler and primarily designed to aggregate all model parameters from every specified instance of ParameterSpace and Discretization.
Parameterization setting that consists of one or more |
|
Utility class to parameterize the Bayesian inference problem |