API Reference#
Configuration#
High-level configuration objects for assembling a calibration problem and handing it to an external sampler (emcee, dynesty, etc.).
End-to-end configuration for Bayesian calibration. |
|
Configuration for a single sector of parameters. |
Priors#
Prior distribution classes that satisfy the generic prior protocol required by
ParameterConfig. Any user-defined class with logpdf,
rvs, and (optionally) prior_transform methods can be used directly.
Independent prior built from an arbitrary list of |
|
Independent truncated-normal prior for a vector of parameters. |
Core building blocks#
Pair observations with a physical model and a likelihood model. |
|
A collection of independent constraints sharing a common physical model. |
|
A class to represent an observation with statistical errors, as well as systematic errors associated with a common normalization and offset of all or some of the data points of the the dependent variable y. |
|
A class to represent an observation with fixed covariance. |
|
A single scalar model parameter. |
|
Abstract base class for parametric physical models. |
|
Polynomial model of fixed order. |
Likelihood models#
A class to represent a likelihood model for comparing an Observation to a PhysicalModel. |
|
A special LikelihoodModel to handle FixedCovarianceObservation objects, where the covariance matrix is fixed and does not depend on the parameters of the PhysicalModel. |
|
A LikelihoodModel that returns the negative half of the chi-squared statistic for the log likelihood, ignoring the log determinant term. |
|
A class to represent a likelihood model for comparing an Observation to a PhysicalModel, in which the LikelihoodModel has it's own parameters to calculate the covariance, aside from the parameters of the PhysicalModel. |
|
A ParametricLikelihoodModel in which each data point in the observation has the same, unknown, statistical error, which is a parameter, $epsilon$. |
|
A ParametricLikelihoodModel in which each data point in the observation has the a statistical error corresponding to a fixed fraction of it's value, the fraction being a parameter, $epsilon$. |
|
A ParametricLikelihoodModel in which the (log of the) multiplicative factor of the model output is a free parameter. |
|
A ParametricLikelihoodModel in which the systematic uncertainty of the normalization of the observation is a parameter, $eta$. |
|
A ParametricLikelihoodModel in which the frac_err is a free parameter $gamma$, such that the covariance due to the uncorrelated model error takes the form: |
|
A LikelihoodModel that uses a Student's t-distribution for the likelihood. |
|
|
Parametric likelihood with a GP discrepancy covariance term. |
Sampling#
Gibbs-style MCMC coordinator for a Bayesian calibration problem. |
|
Base class wrapping a sampling algorithm with chain recording. |
|
Metropolis-Hastings sampler with a fixed proposal distribution. |
|
Metropolis sampler that adapts the proposal covariance with a sliding window. |
|
Metropolis sampler that updates the proposal covariance after each batch. |
|
Abstract base class for Metropolis-Hastings proposal distributions. |
|
Multivariate-normal proposal centred at the current state. |
|
Half-normal proposal, useful for strictly non-negative parameters. |
|
Normal proposal operating in log space, for strictly positive parameters. |
Sampling algorithms#
Low-level sampling functions used internally by the sampler classes.
Metropolis-Hastings MCMC sampling. |
|
Adaptive Metropolis algorithm with sliding-window covariance adaptation. |
Domain-specific models#
Reaction-physics observation and model classes for elastic differential cross sections and isobaric-analog (p,n) cross sections.
|
Observation for elastic differential cross sections. |
A model that predicts the elastic differential xs for a given reaction. |
|
Observation for (p,n) isobaric analog state (IAS) reactions. |
|
A model that predicts the (p,n) IAS differential xs for a given reaction. |