Data and Likelihood

In BayesBay, the observed data should be embedded in a Target. When paired with a forward function that enables data predictions from the considered model parameters, a Target can be utilized to define an instance of LogLikelihood. Importantly, Target facilitates treating the noise properties of the associated data as unknown, enabling hierarchical inversions.

An instance of LogLikelihood can encapsulate multiple Target instances, each associated with a distinct forward function. This enables joint inversions of various types of data sets.

bayesbay.likelihood.Target	Observed data with noise that can be treated as an unknown
bayesbay.likelihood.LogLikelihood	Helper class to evaluate the log likelihood ratio

All examples in this documentation make use of one or more Target instances (and associated forward functions) to initialize LogLikelihood. For usage tips, refer to our Quickstart Tutorial. Examples in this documentation employing hierarchical sampling, where the data noise is treated as unknown, include:

• Polynomial Fitting: Part II

• Gaussian Mixture Modelling: Part I and Part II

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

• Surface-Wave Tomography

• Partition Modelling: Part I and Part II