sheap.MasterSampler.Samplers.MonteCarloSampler module

Monte Carlo Sampler

This module implements the MonteCarloSampler, a simple posterior approximation for spectral fits based on randomized parameter initialization and local re-optimization.

Main Features

• Generates random draws of parameter vectors within their constraints.

• Converts parameters to raw space and re-optimizes them with Minimizer.

• Handles tied/fixed parameters through build_Parameters() and dependency flattening utilities.

• Reconstructs physical parameters from optimized raw vectors.

• Computes physical quantities (fluxes, FWHM, luminosities, etc.) for each draw using SheaProducts.

Public API

• MonteCarloSampler

  • MonteCarloSampler.sample_params() — run the Monte Carlo sampler and return posterior dictionaries.

  • MonteCarloSampler.make_minimizer() — construct a Minimizer configured with penalties/weights.

  • MonteCarloSampler._build_tied() — convert tied-parameter specifications into dependency strings.

Notes

• This method approximates the posterior distribution by repeatedly optimizing from random starts (sometimes called a “poor man’s MCMC”).

• Actual uncertainty propagation is performed by analyzing the distribution of optimized solutions.

• Dependencies are flattened so that all tied parameters ultimately reference free parameters only.

class MonteCarloSampler(estimator)

Bases: object

Montecarlo sampler still under developmen.

Parameters:

estimator (MasterSampler)

sample_params(num_samples=100, key_seed=0, summarize=True, **kwargs)

Parameters:

• num_samples (int)

• key_seed (int)

Return type:

jax.numpy.ndarray

sample_params_st(num_samples=100, key_seed=0, summarize=True, **kwargs)

stable start from the best result.

Parameters:

• num_samples (int)

• key_seed (int)

Return type:

jax.numpy.ndarray

sample_params_experimental(num_samples=100, key_seed=0, summarize=True, return_only_draws=False, frac_box_sigma=0.5, k_sigma=0.5)

Parameters:

• num_samples (int)

• key_seed (int)

Return type:

jax.numpy.ndarray

make_minimizer(model, non_optimize_in_axis, num_steps, learning_rate, method, penalty_weight, curvature_weight, smoothness_weight, max_weight, penalty_function=None, weighted=True, **kwargs)