gwnr.stats — statistics & inference

Bayesian-inference configuration, sampling utilities, Fisher matrices and distribution containers. Source: gwnr/stats/.

1. config_utils.py — configuration infrastructure
2. pycbc_inference_utils.py — PyCBC Inference configs
3. bilby_utils.py — Bilby configs and scripts
4. lal_inference_utils.py — LALInference posteriors
5. samplers.py — emcee helpers
6. sampling.py — drawing from standard distributions
7. distribution.py — posterior containers
8. fisher_information.py — Fisher matrices
9. enigma_utils.py — ENIGMA calibration likelihoods
10. priors.py

---

config_utils.py — configuration infrastructure

• class ConfigWriter — writes stored config-file templates to disk, filling in formatting blanks (e.g. event-specific data options) via write(name, **formatting_kwargs).
• class ConfigBase — stores config-file samples by category as a dict of dicts, and hands out ConfigWriters (available_configs(), get_config_writer(name), get/set).

These are the base classes for the PyCBC Inference and Bilby config generators below.

pycbc_inference_utils.py — PyCBC Inference configs

class InferenceConfigs(ConfigBase) — stores and writes configuration files for pycbc_inference runs: add_data_configs(event_name) (data/PSD options per GW event), add_sampler_configs(n_cpus=10, n_live=2000, n_maxmcmc=8000, d_logz=0.1, n_walkers=1000, n_temperatures=20, ...) (nested-sampling / ensemble-MCMC settings), and add_inference_configs() (model, priors, variable/static parameters). Driven by the gwnr_write_pycbc_inference_configs CLI tool.

bilby_utils.py — Bilby configs and scripts

• class InferenceConfigs(ConfigBase) — Bilby counterpart: default BBH priors, prior files, event and injection configs.
• class BilbyScriptWriterBase and subclasses BilbyScriptWriterInjection / BilbyScriptWriterEvent — programmatically compose complete Bilby run scripts section by section (imports, data/injection setup, template, priors, likelihood, sampler, post-processing) via script_lines(), write_script(), write_prior_file().

lal_inference_utils.py — LALInference posteriors

Utilities for posterior_samples.dat files produced by LALInference:

• get_header_data_from_posterior_samples_file(filename, ...), get_param_idx, get_param_from_names, get_1dslice_posterior, write_posterior_samples_file
• get_h_from_posterior_line(line, header, det_tag, approx='IMRPhenomPv2', ...) — regenerate the waveform corresponding to any posterior sample (no conditioning applied; frequency-domain approximants are not converted to the time domain)
• shift_waveform_phase_time(orig_line, phase_shift, time_shift, sample_rate) — same, with additional phase/time shifts (SI units)

---

samplers.py — emcee helpers

• get_emcee_ensemble_sampler(log_probability, params_to_sample, myarglist, ..., nwalkers=32, burn_in=100, backend_hdf=None, pool=None) — initialize and burn in an emcee ensemble sampler, optionally with an HDF5 checkpoint backend.
• emcee_samples_to_dict(sampler, params_to_sample, burnin=1000, thin=10) — pull thinned samples into a parameter-keyed dictionary.
• emcee_samples_from_checkpoint(checkpoint_file, ...) — same, from a checkpoint file.
• write_output_from_emcee_sampler(output_file_name, sampler, params_to_sample, ...) — dump samples to ASCII.

sampling.py — drawing from standard distributions

Convenience samplers used to build injection sets and priors: uniform_massratio, uniform_mass, uniform_in_totalmass_massratio_masses, uniform_spin_magnitude, uniform_coalescence_time, uniform_in_angle / uniform_in_cos_angle, uniform_on_S2 (isotropic sky/spin directions via cube_to_uniform_on_S2), uniform_in_volume_distance, uniform_in_choices, zero_distribution, idempotence.

class OneDRandom — meta-class mapping named parameters to their distributions (driven by a pandas.DataFrame with a dist column); sample(name, size=1).

distribution.py — posterior containers

• class OneDDistribution — one-dimensional sample sets: mean, median, percentile, kde, normalization, range-restricted variants (mean_in_range, pdf_over_range).
• class MultiDDistribution — n-dimensional sample sets: named columns (index_of_name, sliced), mean/median/percentile/credible_interval per parameter, kde, corner_plot, plot_twod_slice.
• class MultipleOneDDistributions — combine 1-D marginal distributions across many events (e.g. stacking H₀ posteriors): process_oned_slices, combine_oned_slices, plot_combined_oned_slices.

---

fisher_information.py — Fisher matrices

• get_waveform_derivatives_wrt_params(approximant='SEOBNRv2', ..., deriv_params=['mass1','mass2','spin1z','spin2z'], delta_m1=1e-3, ...) — numerical derivatives ∂h/∂θ of the waveform with respect to binary parameters, with adjustable finite-difference steps and tolerance.
• get_correlation_fisher_matrices(..., psd='aLIGOZeroDetHighPower', return_derivs=False, ...) — assembles the Fisher information matrix Γ_ij = (∂h/∂θ_i | ∂h/∂θ_j) and the corresponding correlation (inverse-Fisher) matrix for parameter-uncertainty estimates.

enigma_utils.py — ENIGMA calibration likelihoods

Log-prior/likelihood/probability functions for calibrating attachment parameters of the ENIGMA/ESIGMA eccentric waveform model against reference waveforms: log_prior_esigma, log_likelihood_esigma(mass1, mass2, omega_attach, PNO, f_lower, sample_rate, psd, ...) (match-based likelihood), log_prob_esigma, log_prob_esigma_fixed_masses, log_prob_esigma_fixed_total_mass_hidden_q. Used by the gwnr_enigma_* CLI tools with the emcee helpers above.

The ESIGMA waveform-generation code itself has been moved out of this repository; these calibration utilities remain here.

priors.py

Placeholder module for prior definitions (currently empty).