biogeme.bayesian_estimation.sampling_strategy module

Defines the strategy for sampling the MCMC based on hardware configuration and user’s preferences.

Michel Bierlaire Mon Oct 27 2025, 17:01:13

class biogeme.bayesian_estimation.sampling_strategy.SamplingConfig(backend, chain_method, cores, target_accept, init, max_treedepth, nuts_kwargs)[source]

Bases: object

Parameters:

  • backend (str)

  • chain_method (str | None)

  • cores (int | None)

  • target_accept (float)

  • init (str | None)

  • max_treedepth (int | None)

  • nuts_kwargs (dict[str, Any] | None)

backend: str
chain_method: str | None
cores: int | None
init: str | None
max_treedepth: int | None
nuts_kwargs: dict[str, Any] | None
target_accept: float
biogeme.bayesian_estimation.sampling_strategy.describe_strategies()[source]
Return type:

str

biogeme.bayesian_estimation.sampling_strategy.make_sampling_config(strategy, target_accept)[source]

Create a SamplingConfig from a short strategy string.

Parameters:

  • strategy (str) – One of 'automatic', 'numpyro-parallel', 'numpyro-vectorized', or 'pymc'.

  • target_accept (float) – Target acceptance rate.

Returns:

A ready-to-use configuration object.

Return type:

SamplingConfig

Raises:

ValueError – If strategy is not one of the allowed values.