Builder

abstract_builder

Base class for application builders.

The MTMLDA takes a list of callables as posterior hierarchy. To enhance flexibility w.r.t. to how these models are generated, and to support parallel chains, we employ a builder pattern for applications. With the prescribed interface, the builder can be used by the run wrapper and trigger MTMLDA run from a variety of use-cases.

Classes:

Name	Description
ApplicationBuilder	Base class for application builders.
InverseProblemSettings	Base data class for inverse problem settings.
SamplerComponentSettings	Base data class for sampler component settings.
InitialStateSettings	Base data class for initial state settings.

mtmlda.components.abstract_builder.InverseProblemSettings dataclass

Empty base data class for inverse problem settings.

mtmlda.components.abstract_builder.SamplerComponentSettings dataclass

Empty base data class for sampler component settings.

mtmlda.components.abstract_builder.InitialStateSettings dataclass

Empty base data class for initial state settings.

mtmlda.components.abstract_builder.ApplicationBuilder

Bases: ABC

Base class for application builders.

The class provides a generic interface for application builders, as will be invoked by the run wrapper. The class is purely abstract, implementation must follow in subclasses. The only concrete behavior the class enforces is that the id of the invoking process is provided. This is useful for setting up components in a process-specific way for chain-parallel runs.

Methods:

Name	Description
set_up_models	Set up a model hierarchy in the form of a list of callables..
set_up_sampler_components	Set up proposal and accept rate estimator.
generate_initial_state	Generate the initial state for the Markov chain

__init__

__init__(process_id: int) -> None

Base class constructor, assigns process id of calling process as internal attribute.

Parameters:

Name	Type	Description	Default
process_id	int	ID of calling process	required

set_up_models abstractmethod

set_up_models(inverse_problem_settings: InverseProblemSettings) -> list[Callable]

Set up a posterior hierarchy as a sequence of callables.

Within this method, different components can be freely combined to obtain a posterior hierarchy. For instance, an external model server could be called to provide a likelihood, which is then combined with a prior component. This method is the main reason the separate builder exists, to allow for more flexibility in the model setup.

Parameters:

Name	Type	Description	Default
inverse_problem_settings	InverseProblemSettings	Necessary settings for model setup, can be anything	required

Returns:

Type	Description
list[Callable]	list[Callable]: Posterior hierarchy for MLDA run

set_up_sampler_components abstractmethod

set_up_sampler_components(
    sampler_component_settings: SamplerComponentSettings,
) -> tuple[mcmc.BaseProposal, mcmc.BaseAcceptRateEstimator]

Set up the coarse-level proposal and accept rate estimator for MLDA.

Can be used for process-dependent initialization.

Parameters:

Name	Type	Description	Default
sampler_component_settings	SamplerComponentSettings	Necessary settings for component setup, can be anything	required

Returns:

Type	Description
tuple[mcmc.BaseProposal, mcmc.BaseAcceptRateEstimator]	tuple[mcmc.BaseProposal, mcmc.BaseAcceptRateEstimator]: Proposal and accept rate estimator

generate_initial_state abstractmethod

generate_initial_state(initial_state_settings: InitialStateSettings) -> np.ndarray

Generate initial state of a Markov chain.

Can be used for process-dependent initialization.

Parameters:

Name	Type	Description	Default
initial_state_settings	InitialStateSettings	Necessary settings for initial state setup, can be anything	required

Returns:

Type	Description
np.ndarray	np.ndarray: Initial state