jesterTOV.inference.samplers.jester_sampler.SamplerOutput

jesterTOV.inference.samplers.jester_sampler.SamplerOutput#

class SamplerOutput(samples, log_prob, metadata=<factory>)[source]#

Bases: object

Standardized output from JESTER samplers.

This dataclass provides a uniform interface for accessing samples, log probabilities, and sampler-specific metadata across different sampling backends (FlowMC, SMC, NS-AW).

Variables:

  • samples (dict[str, Array]) – Dictionary of parameter samples. Keys are parameter names, values are JAX arrays of shape (n_samples,) or (n_samples, n_dim). Only contains actual parameters, not metadata fields.

  • log_prob (Array) – Log probability for each sample. Interpretation depends on sampler: - FlowMC/SMC: log posterior probability - NS-AW: log likelihood (nested sampling uses likelihood) Shape: (n_samples,)

  • metadata (dict[str, Any]) – Sampler-specific metadata. Common fields: - FlowMC: {} (empty, MCMC has equal weights) - SMC: {“weights”: Array, “ess”: float} - NS-AW: {“weights”: Array, “logL”: Array, “logL_birth”: Array}

Notes

The log_prob field has different semantics for NS-AW (log likelihood) versus FlowMC/SMC (log posterior). Consumers should check the sampler type when interpreting this field.

__init__(samples, log_prob, metadata=<factory>)#

Methods

__init__(samples, log_prob[, metadata])

Attributes

samples

log_prob

metadata

log_prob: Array#
metadata: dict[str, Any]#
samples: dict[str, Array]#
Contents