Perturbations

Every time an instance of bayesbay.parameterization.Parameterization or bayesbay.LogLikelihood is initialized, functions defining possible model perturbations are generated under the hood using the classes listed below. For example, a uniformly-distributed free parameter is perturbed at a given Markov chain iteration through a function generated by the ParamPerturbation class. Changes in the dimensionality of a trans-dimensional parameterization are managed by BirthPerturbation and DeathPerturbation. In hierarchical sampling, where the standard deviation of the data is treated as unknown, a perturbation function defined in NoisePerturbation is employed.

graph TD; Perturbation-->ParamPerturbation; Perturbation-->BirthPerturbation; Perturbation-->DeathPerturbation; Perturbation-->NoisePerturbation;

bayesbay.perturbations.Perturbation	Base class for a perturbation callable
bayesbay.perturbations.ParamPerturbation	Perturbation on a chosen parameter value
bayesbay.perturbations.BirthPerturbation	Perturbation by creating a new dimension
bayesbay.perturbations.DeathPerturbation	Perturbation by removing an existing dimension
bayesbay.perturbations.NoisePerturbation	Perturbation by changing the noise level estimation