bilby.core.sampler.dynesty3_utils.AcceptanceTrackingRWalk

class bilby.core.sampler.dynesty3_utils.AcceptanceTrackingRWalk(**kwargs)

Bases: EnsembleWalkSampler

This is a modified version of dynesty.sampling.sample_rwalk that runs the MCMC random walk for a user-specified number of a crude approximation to the autocorrelation time.

This is the proposal method used by default for Bilby<2 and corresponds to specifying sample="rwalk"

__init__(**kwargs)

Initialize the internal sampler.

Importantely this sets up the sampler_kwargs that is being passed to each .sample() call

Parameters:

kwargsdict

A dictionary of additional method-specific parameters. This common keywords:

nonboundedarray

Array of boolean values indicating which dimensions are non-bounded.

periodicarray

Array of boolean values indicating which dimensions are periodic.

reflectivearray

Array of boolean values indicating which dimensions are reflective.

ndim: int

Number of dimensions.

__call__(*args, **kwargs)

Call self as a function.

Methods

__init__(**kwargs)	Initialize the internal sampler.
estimate_nmcmc(accept_ratio[, safety, tau, ...])	Estimate autocorrelation length of chain using acceptance fraction
prepare_sampler([loglstar, points, axes, ...])	Prepare the list of arguments for sampling.
sample(args)	Return a new live point proposed by random walking away from an existing live point.
tune(tuning_info[, update])	Update the proposal parameters based on the number of accepted steps and MCMC chain length.

Attributes

citations
old_act
update_bound_interval_ratio	How often to force updating the bounds The value is in units of ncall per nlive.

static estimate_nmcmc(accept_ratio, safety=5, tau=None, maxmcmc=5000, old_act=None)

Estimate autocorrelation length of chain using acceptance fraction

Using ACL = (2/acc) - 1 multiplied by a safety margin. Code adapted from:

• https://github.com/farr/Ensemble.jl

• https://github.com/johnveitch/cpnest/blob/master/cpnest/sampler.py

Parameters:

accept_ratio: float [0, 1]

Ratio of the number of accepted points to the total number of points

old_act: int

The ACT of the last iteration

maxmcmc: int

The maximum length of the MCMC chain to use

safety: int

A safety factor applied in the calculation

tau: int (optional)

The ACT, if given, otherwise estimated.

Notes

This method does not compute a reliable estimate of the autocorrelation length for our proposal distributions.

prepare_sampler(loglstar=None, points=None, axes=None, seeds=None, prior_transform=None, loglikelihood=None, nested_sampler=None)

Prepare the list of arguments for sampling.

Parameters:

loglstarfloat

Ln(likelihood) bound.

points~numpy.ndarray with shape (n, ndim)

Initial sample points.

axes~numpy.ndarray with shape (ndim, ndim)

Axes used to propose new points.

seeds~numpy.ndarray with shape (n,)

Random number generator seeds.

prior_transformfunction

Function transforming a sample from the a unit cube to the parameter space of interest according to the prior.

loglikelihoodfunction

Function returning ln(likelihood) given parameters as a 1-d ~numpy array of length ndim.

nested_sampler~dynesty.samplers.Sampler

The nested sampler object used to sample.

Returns:

arglist:

List of SamplerArgument objects containing the parameters needed for sampling.

static sample(args)

Return a new live point proposed by random walking away from an existing live point.

Parameters:

u~numpy.ndarray with shape (ndim,)

Position of the initial sample. This is a copy of an existing live point.

loglstarfloat

Ln(likelihood) bound.

axes~numpy.ndarray with shape (ndim, ndim)

Axes used to propose new points. For random walks new positions are proposed using the Ellipsoid whose shape is defined by axes.

scalefloat

Value used to scale the provided axes.

prior_transformfunction

Function transforming a sample from the a unit cube to the parameter space of interest according to the prior.

loglikelihoodfunction

Function returning ln(likelihood) given parameters as a 1-d ~numpy array of length ndim.

kwargsdict

A dictionary of additional method-specific parameters.

Returns:

u~numpy.ndarray with shape (ndim,)

Position of the final proposed point within the unit cube.

v~numpy.ndarray with shape (ndim,)

Position of the final proposed point in the target parameter space.

loglfloat

Ln(likelihood) of the final proposed point.

ncint

Number of function calls used to generate the sample.

sampling_infodict

Collection of ancillary quantities used to tune scale.

tune(tuning_info, update=True)

Update the proposal parameters based on the number of accepted steps and MCMC chain length.

The walks parameter to asymptotically approach the desired number of accepted steps.

property update_bound_interval_ratio

How often to force updating the bounds The value is in units of ncall per nlive. I.e. the value of 10 means for N live points, the bound will be updated every 10 * N calls