nessai.flowsampler

Main code that handles running and checkpoiting the sampler.

Module Contents

Classes

FlowSampler

Main class to handle running the nested sampler.

class nessai.flowsampler.FlowSampler(model, output=os.getcwd(), importance_nested_sampler=False, resume=True, resume_file='nested_sampler_resume.pkl', resume_data=None, weights_file=None, weights_path=None, signal_handling=True, exit_code=130, pytorch_threads=1, close_pool=True, eps=None, torch_dtype=None, disable_vectorisation=False, likelihood_chunksize=None, allow_multi_valued_likelihood=None, parallelise_prior=None, result_extension='hdf5', **kwargs)

Main class to handle running the nested sampler.

Parameters:

modelnessai.model.Model

User-defined model.

outputstr, optional

Output directory

resumebool, optional

If True try to resume the sampler is the resume file exists.

resume_filestr, optional

File to resume sampler from.

resume_data: Any, optional

Data to resume the sampler from instead of a resume file. The data will be passed to the resume_from_pickled_sampler of the relevant class.

weights_pathstr, optional

Path to either the weights file or directory containing subdirectories with weight files.

weights_filestr, optional

Weights file to use, same as weights_path and included for backwards_compatibility. Will be ignored is weights_paths has also been specified.

pytorch_threadsint

Maximum number of threads to use for torch. If None torch uses all available threads.

signal_handlingbool

Enable or disable signal handling.

exit_codeint, optional

Exit code to use when forcibly exiting the sampler.

close_poolbool

If True, the multiprocessing pool will be closed once the run method has been called. Disables the option in NestedSampler if enabled.

disable_vectorisationbool

Disable likelihood vectorisation. Overrides the value of nessai.model.Model.allow_vectorised.

likelihood_chunksizeOptional[int]

Chunksize used when evaluating a vectorised likelihood. Overrides the of nessai.model.Model.likelihood_chunksize. Set to None to evaluate the likelihood with all available points.

allow_multi_valued_likelihoodOptional[bool]

Allow for a multi-valued likelihood function that will return different likelihood values for the same point in parameter space. See nessai.model.Model.allow_multi_valued_likelihood for more details.

parallelise_priorOptional[bool]

If true, and a multiprocessing pool has been specified, then the log-prior calculation will be parallelised using multiprocessing. If false, then the pool with not be used. Overrides the value of the parallelise_prior attribute in the model class, which is false by default.

result_extensionstr

Extension used when saving the result format. Defaults to HDF5, but also supports JSON.

kwargs

Keyword arguments passed to NestedSampler.

property log_evidence

Return the most recent log evidence

property log_evidence_error

Return the most recent log evidence error

property nested_samples

Return the nested samples

check_resume(resume_file, resume_data)

Check if it is possible to resume the sampler

run(plot=True, save=True, posterior_sampling_method=None, close_pool=None, **kwargs)

Run the nested sampler.

Will pick the correct run method given the configuration used.

Parameters:

plotbool

Toggle all plots produced once the sampler has converged.

savebool, optional

Toggle automatic saving of results

posterior_sampling_methodstr, optional

Method used for drawing posterior samples. Defaults to rejection sampling.

close_poolbool, optional

Boolean to indicated if the pool should be closed at the end of the run function. If False, the user must manually close the pool. If specified, this value overrides the value passed when initialising the FlowSampler class.

run_standard_sampler(plot=True, plot_indices=True, plot_posterior=True, plot_logXlogL=True, save=True, posterior_sampling_method=None, close_pool=None)

Run the standard nested sampler.

Parameters:

plotbool

Toggle all plots produced once the sampler has converged.

plot_indicesbool

Toggle the insertion indices plot.

plot_posteriorbool

Toggle the posterior distribution plot.

plot_logXlogLbool

Toggle the log-prior volume vs log-likelihood plot.

save

Enable or disable saving of a results file.

posterior_sampling_method

Method used for drawing posterior samples. Defaults to rejection sampling.

close_poolbool

Boolean to indicated if the pool should be closed at the end of the run function. If False, the user must manually close the pool. If specified, this value overrides the value passed when initialising the class.

run_importance_nested_sampler(plot=True, plot_posterior=True, save=True, posterior_sampling_method=None, redraw_samples=False, n_posterior_samples=None, compute_initial_posterior=False, close_pool=None, **kwargs)

Run the importance nested sampler.

Parameters:

plot

Enable or disable plotting. Independent of the value passed to the NestedSampler object.

save

Enable or disable saving of a results file.

posterior_sampling_method

Method used for drawing posterior samples. Defaults to importance sampling.

redraw_samples

If True after the sampling is finished, samples are redrawn from the meta proposal and used to compute an updated evidence estimate and posterior. This can reduce biases in the results.

n_posterior_samples

Number of posterior samples to draw when when redrawing samples.

compute_initial_posterior

Enables or disables computing the posterior before redrawing samples. If redraw_samples is False, then this flag is ignored.

close_poolbool

Boolean to indicated if the pool should be closed at the end of the run function. If False, the user must manually close the pool. If specified, this value overrides the value passed when initialising the class.

kwargs

Keyword arguments passed to draw_final_samples()

save_kwargs(kwargs: dict) → None

Save the dictionary of keyword arguments used.

Uses an encoder class to handle numpy arrays.

Parameters:

kwargsdict

Dictionary of kwargs to save.

save_results(filename: str, extension: str | None = None) → None

Save the results from sampling to a specific results file.

Parameters:

filenamestr

Name of file to save results to.

extensionOptional[str]

File extension to used. If not specified, it will be inferred from the filename.

terminate_run(code=None)

Terminate a sampling run.

Parameters:

codeint, optional

Code passed to close_pool

safe_exit(signum=None, frame=None)

Safely exit. This includes closing the multiprocessing pool.