Data Analysis

Module to run parallel bilby using MPI.

Command line interface for data analysis

usage: parallel_bilby_analysis [-h] [--version] [-n NLIVE] [--dlogz DLOGZ]
                               [--n-effective N_EFFECTIVE]
                               [--dynesty-sample DYNESTY_SAMPLE]
                               [--dynesty-bound DYNESTY_BOUND] [--walks WALKS]
                               [--proposals PROPOSALS] [--maxmcmc MAXMCMC]
                               [--nact NACT] [--naccept NACCEPT]
                               [--min-eff MIN_EFF] [--facc FACC]
                               [--enlarge ENLARGE]
                               [--n-check-point N_CHECK_POINT]
                               [--max-its MAX_ITS]
                               [--max-run-time MAX_RUN_TIME]
                               [--fast-mpi FAST_MPI] [--mpi-timing MPI_TIMING]
                               [--mpi-timing-interval MPI_TIMING_INTERVAL]
                               [--nestcheck]
                               [--rejection-sample-posterior REJECTION_SAMPLE_POSTERIOR]
                               [--bilby-zero-likelihood-mode]
                               [--sampling-seed SAMPLING_SEED] [-c]
                               [--no-plot] [--do-not-save-bounds-in-resume]
                               [--check-point-deltaT CHECK_POINT_DELTAT]
                               [--rotate-checkpoints] [--outdir OUTDIR]
                               [--label LABEL] [--result-format RESULT_FORMAT]
                               data_dump

Positional Arguments

data_dump

The pickled data dump generated by parallel_bilby_analysis

Named Arguments

--version

show program’s version number and exit

--outdir

Outdir to overwrite input label

--label

Label to overwrite input label

--result-format

Format to save the result

Default: “hdf5”

Dynesty Settings

-n, --nlive

Number of live points

Default: 1000

--dlogz

Stopping criteria: remaining evidence, (default=0.1)

Default: 0.1

--n-effective

Stopping criteria: effective number of samples, (default=inf)

Default: inf

--dynesty-sample

Dynesty sampling method (default=acceptance-walk). Note, the dynesty rwalk method is overwritten by parallel bilby for an optimised version

Default: “acceptance-walk”

--dynesty-bound

Dynesty bounding method (default=multi)

Default: “live”

--walks

Minimum number of walks, defaults to 100

Default: 100

--proposals

The jump proposals to use, the options are ‘diff’ and ‘volumetric’

--maxmcmc

Maximum number of walks, defaults to 5000

Default: 5000

--nact

Number of autocorrelation times to take, defaults to 2

Default: 2

--naccept

The average number of accepted steps per MCMC chain, defaults to 60

Default: 60

--min-eff

The minimum efficiency at which to switch from uniform sampling.

Default: 10

--facc

See dynesty.NestedSampler

Default: 0.5

--enlarge

See dynesty.NestedSampler

Default: 1.5

--n-check-point

Steps to take before attempting checkpoint

Default: 1000

--max-its

Maximum number of iterations to sample for (default=1.e10)

Default: 10000000000

--max-run-time

Maximum time to run for (default=1.e10 s)

Default: 10000000000.0

--fast-mpi

Fast MPI communication pattern (default=False)

Default: False

--mpi-timing

Print MPI timing when finished (default=False)

Default: False

--mpi-timing-interval

Interval to write timing snapshot to disk (default=0 – disabled)

Default: 0

--nestcheck

Save a ‘nestcheck’ pickle in the outdir (default=False). This pickle stores a nestcheck.data_processing.process_dynesty_run object, which can be used during post processing to compute the implementation and bootstrap errors explained by Higson et al (2018) in “Sampling Errors In Nested Sampling Parameter Estimation”.

Default: False

--rejection-sample-posterior

Whether to generate the posterior samples by rejection sampling the nested samples or resampling with replacement

Default: True

Misc. Settings

--bilby-zero-likelihood-mode

Default: False

--sampling-seed

Random seed for sampling, parallel runs will be incremented

-c, --clean

Run clean: ignore any resume files

Default: False

--no-plot

If true, don’t generate check-point plots

Default: False

--do-not-save-bounds-in-resume

If true, do not store bounds in the resume file. This can make resume files large (~GB)

Default: True

--check-point-deltaT

Write a checkpoint resume file and diagnostic plots every deltaT [s]. Default: 1 hour.

Default: 3600

--rotate-checkpoints

If true, backup checkpoint before overwriting (ending in ‘.bk’).

Default: False