CalibrationProfile

An immutable record of the PDF and entropy chunk budgets used by a run. Obtain a measured profile with calibrate, or construct one explicitly to reproduce chosen settings. Profiles carry compact provenance; detailed campaign measurements are kept separately in the cache.

from pyvbmc import CalibrationProfile, VariationalPosterior

profile = CalibrationProfile(
    pdf_chunk_elements=65536,
    entropy_grad_chunk_elements=65536,
    entropy_value_chunk_elements=65536,
)
vp = VariationalPosterior(D=4, K=20, calibration=profile)

The budgets stay fixed for the lifetime of a resolved posterior. Copies, saved posteriors and resumed VBMC runs preserve them. Machine compatibility controls automatic cache reuse; it does not replace an explicitly supplied or restored profile.

class pyvbmc.CalibrationProfile(pdf_chunk_elements: int = 65536, entropy_grad_chunk_elements: int = 65536, entropy_value_chunk_elements: int = 65536, source: str = 'explicit', status: str = 'complete', fingerprint: str | None = None, provenance: Mapping[str, Any] = {}, cache_path: str | None = None)

Fixed chunk settings selected for one PyVBMC run.

The profile stores only compact provenance. Detailed measurements belong to the machine-local calibration report rather than posterior histories.

Parameters:

pdf_chunk_elementsint, optional

Element budget for posterior density evaluation.

entropy_grad_chunk_elementsint, optional

Element budget for Monte Carlo entropy calls that request gradients.

entropy_value_chunk_elementsint, optional

Element budget for value-only Monte Carlo entropy calls.

sourcestr, optional

Provenance label for the settings. Profiles returned by calibrate() use "calibrated", "memory", "cache" or "default". The default for a directly constructed profile is "explicit".

statusstr, optional

Status associated with the settings. Profiles returned by calibrate() use "complete", "busy", "incomplete" or "invalid". The default is "complete".

fingerprintstr or None, optional

Compatible machine/configuration fingerprint, when available.

provenancemapping, optional

Compact string-keyed metadata with JSON scalar values.

cache_pathstr or None, optional

Calibration report path, when the profile was persisted.

cache_path: str | None = None

entropy_grad_chunk_elements: int = 65536

entropy_value_chunk_elements: int = 65536

fingerprint: str | None = None

classmethod from_dict(data: Mapping[str, Any]) → CalibrationProfile

Validate and deserialize a profile from its versioned schema.

pdf_chunk_elements: int = 65536

provenance: Mapping[str, Any] = {}

schema_version: ClassVar[int] = 1

property settings: dict[str, int]

Return the three numerical settings as a new dictionary.

source: str = 'explicit'

status: str = 'complete'

to_dict() → dict[str, Any]

Serialize this profile to its versioned, JSON-compatible schema.

property uses_historical_defaults: bool

Whether all numerical settings equal the historical constants.