bilby.bilby_mcmc.proposals.FisherMatrixProposal

class bilby.bilby_mcmc.proposals.FisherMatrixProposal(priors, subset=None, weight=1, update_interval=100, scale_init=1.0, fd_eps=0.0001, adapt=False)

Bases: AdaptiveGaussianProposal

__init__(priors, subset=None, weight=1, update_interval=100, scale_init=1.0, fd_eps=0.0001, adapt=False)

__call__(chain, likelihood=None, priors=None)

Call self as a function.

Methods

__init__(priors[, subset, weight, ...])
apply_boundaries(point)
apply_periodic_boundary(key, val)
apply_reflective_boundary(key, val)
check_dependencies([warn])	Check the dependencies required to use the proposal
propose(chain, likelihood, priors)	Propose a new point
update_scale(chain)	The adaptation of the scale follows (35)/(36) of https://arxiv.org/abs/1409.7215

Attributes

acceptance_ratio
accepted
n
needs_likelihood_and_priors	Fisher Matrix Proposals
rejected

static check_dependencies(warn=True)

Check the dependencies required to use the proposal

Parameters:

warn: bool

If true, print a warning

Returns:

check: bool

If true, dependencies exist

needs_likelihood_and_priors = True

Fisher Matrix Proposals

Uses a finite differencing approach motivated by BayesWave (see, e.g. https://arxiv.org/abs/1410.3835). The inverse Fisher Information Matrix is calculated from the current sample, then proposals are drawn from a multivariate Gaussian and scaled by an adaptive parameter.

propose(chain, likelihood, priors)

Propose a new point

This method must be overwritten by implemented proposals. The propose method is called by __call__, then boundaries applied, before returning the proposed point.

Parameters:

chain: bilby.core.sampler.bilby_mcmc.chain.Chain

The chain to use for the proposal

Returns:

proposal: bilby.core.sampler.bilby_mcmc.Sample

The proposed point

log_factor: float

The natural-log of the additional factor entering the acceptance probability to ensure detailed balance. For symmetric proposals, a value of 0 should be returned.

update_scale(chain)

The adaptation of the scale follows (35)/(36) of https://arxiv.org/abs/1409.7215