sheap.MasterSampler.Samplers.McMcSampler module

MCMC Sampler (NumPyro)

This module provides the McMcSampler, a wrapper around numpyro.infer.MCMC + NUTS for sampling posterior distributions of spectral fit parameters.

Interfaces directly with a MasterSampler estimator (after a fit has been run).
Prepares normalized spectra, constraints, and parameter dictionaries for NumPyro.
Builds a model function via make_numpyro_model().
Runs Hamiltonian Monte Carlo (No-U-Turn Sampler).
Reconstructs full parameter vectors from sampled free parameters, applying tied and fixed constraints.
Rescales amplitude/log-amplitude parameters back into original units.
Wraps posterior samples into physical quantities using SheaProducts.

McMcSampler:
- McMcSampler.sample_params() — run the sampler for one or more spectra, returning posterior parameter dictionaries.

Notes

Dependencies (ties/fixes) are enforced via sheap.Assistants.parser_mapper.apply_tied_and_fixed_params().
By default, each parameter is renamed to theta_N for NumPyro’s sampler to avoid issues with long names.
Internally uses JAX PRNG keys; n_random and key_seed can be used to control reproducibility.

class McMcSampler(estimator)[source]

Bases: object

sample_params(num_samples=2000, num_warmup=500, summarize=True, n_random=1000, list_of_objects=None, key_seed=0)[source]

Parameters: