cogwheel.sampling.Sampler¶

class cogwheel.sampling.Sampler(posterior, run_kwargs=None, sample_prior=False, dir_permissions=493, file_permissions=420)¶

Bases: ABC, JSONMixin

Generic base class for sampling distributions.

Subclasses implement the interface with specific sampling codes.

Parameter space folding is used; this means that the some (“folded”) dimensions are sampled over half their original range, and a map to the other half of the range is defined by reflecting or shifting about the midpoint. The folded posterior distribution is defined as the sum of the original posterior over all 2**n_folds mapped points. This is intended to reduce the number of modes in the posterior.

Parameters:

posteriorcogwheel.posterior.Posterior: Implements the prior and likelihood.
run_kwargsdict: Keyword arguments for the sampler or its run method. Allowed keys depend on the particular sampler used.
sample_priorFalse: Deprecated, will raise ValueError if it is not False.
dir_permissions, file_permissionsoctal: Directory and file permissions.

Methods

`completed`	Return whether the run completed successfully.
`get_init_dict`	Keyword arguments to instantiate the class.
`get_module_name`	Name of the module that defines the instance's class.
`get_rundir`	Return a pathlib.Path object with a new run directory, following a standardized naming scheme for output directories.
`load_evidence`	Define for sampling classes which compute evidence.
`load_samples`	Return a pandas.DataFrame with the posterior samples.
`reinstantiate`	Return a new instance of the class, possibly updating init_kwargs.
`run`	Make a directory to save results and run sampler.
`run_kwargs_options`	Return list of possible parameters to configure sampler, along with their default values.
`submit_condor`	Submit a parameter estimation run using the HTCondor scheduler.
`submit_lsf`
`submit_slurm`
`to_json`	Write class instance to json file.

Attributes

`DEFAULT_RUN_KWARGS`
`JSON_FILENAME`
`PROFILING_FILENAME`
`sampled_params`	Like `.posterior.prior.sampled_params` but the folded parameters have 'folded_' prepended.
`subclass_registry`

static completed(rundir) → bool¶: Return whether the run completed successfully.

get_init_dict(**kwargs)¶: Keyword arguments to instantiate the class.

get_module_name()¶: Name of the module that defines the instance’s class.

get_rundir(parentdir)¶

Return a pathlib.Path object with a new run directory, following a standardized naming scheme for output directories. Directory will be of the form {parentdir}/{prior_name}/{eventname}/{RUNDIR_PREFIX}{run_id}

Parameters:

parentdiros.PathLike: Path to a directory where to store parameter estimation data.

load_evidence() → dict¶

Define for sampling classes which compute evidence.

Return a dict with the following items:

‘log_ev’ = log evidence from sampling
‘log_ev_std’ = log standard deviation of evidence

If using nested importance sampling (NIS), should also have:

‘log_ev_NIS’ = log evidence from nested importance sampling
‘log_ev_std_NIS’ = log standard deviation of NIS evidence

load_samples()¶: Return a pandas.DataFrame with the posterior samples.

reinstantiate(**new_init_kwargs)¶

Return a new instance of the class, possibly updating init_kwargs.

Values not passed will be taken from the current instance.

run(rundir)¶

Make a directory to save results and run sampler.

Parameters:

rundiros.PathLike: Directory where to save output, will create if needed.

run_kwargs_options()¶

Return list of possible parameters to configure sampler, along with their default values.

Refer to each sampler’s documentation for details. Sampler parameters not listed here are handled automatically by cogwheel.

Returns:

list of inspect.Parameter

property sampled_params¶

Like .posterior.prior.sampled_params but the folded parameters have ‘folded_’ prepended.

Returns:

list of str

submit_condor(rundir, request_cpus=1, request_memory='8G', request_disk='1G', resuming=False, postprocess=True, **submit_kwargs)¶

Submit a parameter estimation run using the HTCondor scheduler.

This method generates ‘submit.sub’ and ‘executable.sh’ files, the user should provide any instructions for the submit file as **submit_kwargs.

Parameters:

rundirstr, os.PathLike: Run directory, e.g. from self.get_rundir
request_cpus, request_memory, request_diskint or str: Specifications in the HTCondor submit file.
resumingbool: Whether to attempt resuming a previous run if rundir already exists.
postprocessbool: Whether to perform convergence tests to the run after sampling. See postprocessing.postprocess_rundir.
**submit_kwargs: Further options to include in the HTCondor submit file. Do not pass executable, output, error, log, queue, which will be dealt with automatically.

submit_lsf(rundir, n_hours_limit=48, memory_per_task='32GB', resuming=False, bsub_cmds=(), postprocess=True)¶

Parameters:

rundirstr, os.PathLike: Run directory, e.g. from self.get_rundir
n_hours_limitint: Number of hours to allocate for the job.
memory_per_taskstr: Determines the memory and number of cpus.
resumingbool: Whether to attempt resuming a previous run if rundir already exists.
bsub_cmdstuple of str: Strings with BSUB commands.
postprocessbool: Whether to perform convergence tests to the run after sampling. See postprocessing.postprocess_rundir.

submit_slurm(rundir, n_hours_limit=48, memory_per_task='32G', resuming=False, sbatch_cmds=(), postprocess=True)¶

Parameters:

rundirstr, os.PathLike: Run directory, e.g. from self.get_rundir
n_hours_limitint: Number of hours to allocate for the job.
memory_per_taskstr: Determines the memory and number of cpus.
resumingbool: Whether to attempt resuming a previous run if rundir already exists.
sbatch_cmdstuple of str: Strings with SBATCH commands.
postprocessbool: Whether to perform convergence tests to the run after sampling. See postprocessing.postprocess_rundir.

to_json(dirname, basename=None, *, dir_permissions=493, file_permissions=420, overwrite=False)¶

Write class instance to json file.

It can then be loaded with read_json.