Parameterization

In BayesBay, all free parameters of an inference problem (each characterized by a prior probability, see Prior) should be encapsulated within one or more instances of ParameterSpace (or, alternatively, of Discretization). 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 iteration.

One or more instances of ParameterSpace allow for defining a Parameterization. Compared to ParameterSpace, the Parameterization object is simpler and primarily designed to aggregate all model parameters from every specified instance of ParameterSpace and Discretization.

bayesbay.parameterization.Parameterization	Parameterization setting that consists of one or more ParameterSpace instances
bayesbay.parameterization.ParameterSpace	Utility class to parameterize the Bayesian inference problem

All examples in this documentation involve the use of Parameterization. Examples using ParameterSpace include:

• Quickstart Tutorial

• Polynomial Fitting: Part I and Part II

• Gaussian Mixture Modelling: Part I and Part II

Examples of trans-dimensional parameterizations can be found in:

• Gaussian Mixture Modelling: Part II

• Inversion of Surface-Wave Dispersion Curves: Part I and Part II

• Surface-Wave Tomography

• Partition Modelling: Part II