pastas.solver.mcmc.Emcee#

class pastas.solver.mcmc.Emcee(model: pastas.typing.Model, name: str = 'solver', objfunction: pastas.solver.objective.likelihood.LikelihoodBase | None = None, nwalkers: int = 20, nsteps: int = 5000, backend: Any | None = None, moves: Any | collections.abc.Iterable[Any] | None = None, parallel: bool | None = None, progress_bar: bool = True, **kwargs: Any)#

Solver based on MCMC approach in emcee [Foreman-Mackey et al., 2013].

Parameters:

  • objfunction (pastas.solver.objective.likelihood.LikelihoodBase, optional) – The objective function to be minimized.

  • nwalkers (int, optional) – Number of walkers to use. Default is 20.

  • backend (emcee.backend, optional) – One of the Backends from Emcee used to store MCMC results. See the Emcee documentation for more information.

  • moves (emcee.moves, optional) – The moves argument determines how the next step for a walker is chosen in the MCMC approach. One of the Moves classes from Emcee has to be provided. See Emcee documentation for more information.

  • parallel (bool, optional) – Run the sampler in parallel or not. By default, the parallel option is set to the value of the global Pastas option.

  • progress_bar (bool, optional) – Show the progress bar or not. Requires the tqdm package to be installed.

  • **kwargs – All other keyword arguments are passed on to the SolverBase class.

  • optional – All other keyword arguments are passed on to the SolverBase class.

See also

emcee.EnsembleSampler, emcee.moves, emcee.backend, pastas.solver.objective.likelihood

Notes

The EmceeSolve solver uses the emcee package to perform a Markov Chain Monte Carlo (MCMC) approach to find the optimal parameter values. The solver can be used as follows:

solver = ps.solver.Emcee(ml, nwalkers=20, nsteps=5000, progress_bar=True)

The attributes provided are mostly passed on to the emcee.EnsembleSampler and determine how that instance is created. Keyword arguments to model.solve are parsed to emcee.EnsembleSampler.sample via the run_mcmc method like:

ml.solve(thin_by=2)

To obtain the MCMC chains, use:

ml.solver.sampler.get_chain(flat=True, discard=3000)

References

https://emcee.readthedocs.io/en/stable/

Methods#

get_init_parameters(→ pandas.DataFrame)

Get initial parameters for the solver.

solve(→ tuple[bool, pandas.DataFrame])

Solve the model using MCMC.

log_probability(→ float)

Full log-probability called by Emcee.

log_likelihood(→ float)

Log-likelihood function.

log_prior(→ float)

Probability of parameter set given the priors.

fit_report(→ str)

Report on the fit after a model is optimized.

to_dict(→ dict)

Not supported for this solver.