Skip to content

Backend

Generic backend config

BackendConfig pydantic-model

Bases: BaseModel

Backend config models with shared tracing options.

Config:

  • default: MODEL_CONFIG

Fields:

plot_every = 50 pydantic-field

Write the first progress plot after this many evaluations. Later plots follow plot_step_percent on a log-scaled schedule.

plot_step_percent = 10 pydantic-field

Relative plotting cadence after the first plot. The default of 10 reproduces the current 10%-per-decade schedule.

trace = True pydantic-field

Whether to actively trace the run or not.

validate_for_solver(solver)

Validate solver-dependent config constraints.

Available backends

Nautilus pydantic-model

Bases: BackendConfig

Configuration for the nautilus backend.

Fields:

n_batch = 100 pydantic-field

Minimal number of likelihood evaluation performed at once.

num_live_points = 3000 pydantic-field

Number of live points used by the nested sampler.

plot_every = 50 pydantic-field

Write the first progress plot after this many evaluations. Later plots follow plot_step_percent on a log-scaled schedule.

plot_step_percent = 10 pydantic-field

Relative plotting cadence after the first plot. The default of 10 reproduces the current 10%-per-decade schedule.

trace = True pydantic-field

Whether to actively trace the run or not.

validate_for_solver(solver)

Validate solver-dependent config constraints.

Nessai pydantic-model

Bases: BackendConfig

Configuration for the nessai backend.

Fields:

Validators:

  • _validate_engine_key_conflicts

engine_kwargs pydantic-field

Additional keyword arguments forwarded to nessai's FlowSampler. Reserved keys such as 'nlive', 'output', and 'importance_nested_sampler' are not allowed.

num_live_points = 3000 pydantic-field

Number of live points used by the nested sampler.

plot_every = 50 pydantic-field

Write the first progress plot after this many evaluations. Later plots follow plot_step_percent on a log-scaled schedule.

plot_step_percent = 10 pydantic-field

Relative plotting cadence after the first plot. The default of 10 reproduces the current 10%-per-decade schedule.

trace = True pydantic-field

Whether to actively trace the run or not.

validate_for_solver(solver)

Validate solver-dependent config constraints.

Ultranest pydantic-model

Bases: BackendConfig

Configuration for the ultranest backend.

Fields:

num_live_points = 1000 pydantic-field

Number of live points used by the nested sampler.

plot_every = 50 pydantic-field

Write the first progress plot after this many evaluations. Later plots follow plot_step_percent on a log-scaled schedule.

plot_step_percent = 10 pydantic-field

Relative plotting cadence after the first plot. The default of 10 reproduces the current 10%-per-decade schedule.

resume = True pydantic-field

Resume a previous UltraNest run if checkpoint files exist.

trace = True pydantic-field

Whether to actively trace the run or not.

use_step_sampler = False pydantic-field

Enable UltraNest's population step sampler.

validate_for_solver(solver)

Validate solver-dependent config constraints.

LevenbergMarquardt pydantic-model

Bases: BackendConfig

Configuration for the Levenberg-Marquardt backend.

Fields:

Validators:

  • _validate_x0_physical → x0_physical
  • _validate_initialisation_strategy

engine_kwargs pydantic-field

Additional keyword arguments forwarded to scipy.optimize.least_squares().

init_from_prior = False pydantic-field

Initialise the optimizer from a draw from the prior.

plot_every = 50 pydantic-field

Write the first progress plot after this many evaluations. Later plots follow plot_step_percent on a log-scaled schedule.

plot_step_percent = 10 pydantic-field

Relative plotting cadence after the first plot. The default of 10 reproduces the current 10%-per-decade schedule.

reparametrise = False pydantic-field

If True, search in unconstrained logit-of-CDF space. If False (default), search in physical space with explicit prior bounds via scipy's trf method — matches XSPEC's leven.

trace = True pydantic-field

Whether to actively trace the run or not.

x0_physical = None pydantic-field

Initial parameter vector in physical parameter space. Must contain one value per free parameter.

Iminuit pydantic-model

Bases: BackendConfig

Configuration for the iminuit backend.

Fields:

Validators:

  • _validate_x0_physical → x0_physical
  • _validate_initialisation_strategy

engine_kwargs pydantic-field

Additional keyword arguments forwarded to Minuit.migrad().

init_from_prior = False pydantic-field

Initialise the optimizer from a draw from the prior.

plot_every = 50 pydantic-field

Write the first progress plot after this many evaluations. Later plots follow plot_step_percent on a log-scaled schedule.

plot_step_percent = 10 pydantic-field

Relative plotting cadence after the first plot. The default of 10 reproduces the current 10%-per-decade schedule.

run_hesse = True pydantic-field

Run HESSE after MIGRAD to estimate parameter covariance.

tol = None pydantic-field

MIGRAD EDM stopping tolerance. The actual stop threshold is 0.002 * tol * errordef (so iminuit's default tol=0.1 with errordef=1 gives an EDM threshold of 2e-4). Set to None to keep iminuit's default.

trace = True pydantic-field

Whether to actively trace the run or not.

x0_physical = None pydantic-field

Initial parameter vector in physical parameter space. Must contain one value per free parameter.

Emcee pydantic-model

Bases: BackendConfig

Configuration for the emcee backend.

Fields:

Validators:

  • _validate_x0_physical → x0_physical
  • _validate_initialisation_strategy

init_from_prior = False pydantic-field

Initialise the optimizer from a draw from the prior.

init_spread = 0.001 pydantic-field

Relative Gaussian spread applied to the initial point to generate the walker ensemble.

num_samples = 10000 pydantic-field

Number of production samples to draw after warmup.

num_walkers = 32 pydantic-field

Number of walkers in the ensemble sampler.

num_warmup = 100 pydantic-field

Number of warmup or burn-in steps to discard.

plot_every = 50 pydantic-field

Write the first progress plot after this many evaluations. Later plots follow plot_step_percent on a log-scaled schedule.

plot_step_percent = 10 pydantic-field

Relative plotting cadence after the first plot. The default of 10 reproduces the current 10%-per-decade schedule.

trace = True pydantic-field

Whether to actively trace the run or not.

x0_physical = None pydantic-field

Initial parameter vector in physical parameter space. Must contain one value per free parameter.