descriptors

Computation descriptor abstractions.

Provides FitterDescriptor and EvaluatorDescriptor for declaring metadata about fitters and evaluators.

class pysatl_core.distributions.computations.descriptors.EvaluatorDescriptor(name, target, sources, characteristic_options=(), computation_options=(), constraint_tags=<factory>, description='', evaluator=None)[source]

Bases: _BaseDescriptor

Complete metadata for a non-cacheable evaluator.

An evaluator is lightweight and called on every query without caching. It returns the computed value directly rather than a FittedComputationMethod.

Parameters:

  • name (str) – Unique human-readable identifier.

  • target (str) – Characteristic produced by this evaluator.

  • sources (Sequence[str]) – Characteristics consumed by this evaluator (typically length 1).

  • evaluator (Callable[..., GenericAlias[floating[Any] | integer[Any]]]) – The actual evaluator callable.

  • characteristic_options (tuple[CharacteristicOption, ...]) – Options intrinsic to the characteristic (shared with fitters). These affect the meaning of the result.

  • computation_options (tuple[ComputationOption, ...]) – Options controlling the numerical algorithm used on every call. Unlike fitter computation options (used once at fit-time), evaluator computation options are applied on each invocation. Examples: integration tolerance, finite-difference step, iteration limit.

  • constraint_tags (frozenset[str]) – Constraint tags used for matching.

  • description (str) – Human-readable summary of what the evaluator does.

evaluator: EvaluatorFunc
to_computation_method()[source]

Build an EvaluatorMethod (computation method) from this descriptor.

Return type:

EvaluatorMethod

__init__(name, target, sources, characteristic_options=(), computation_options=(), constraint_tags=<factory>, description='', evaluator=None)
Parameters:
Return type:

None

class pysatl_core.distributions.computations.descriptors.FitterDescriptor(name, target, sources, characteristic_options=(), computation_options=(), constraint_tags=<factory>, description='', fitter=None)[source]

Bases: _BaseDescriptor

Complete metadata for a cacheable fitter.

A fitter performs expensive precomputation and returns a FittedComputationMethod that can be cached and reused.

Parameters:

  • name (str) – Unique human-readable identifier (e.g. "pdf_to_cdf_1C").

  • target (str) – Characteristic produced by this fitter.

  • sources (Sequence[str]) – Characteristics consumed by this fitter (typically length 1).

  • fitter (Callable[..., Any]) – The actual fitting callable.

  • characteristic_options (tuple[CharacteristicOption, ...]) – Options intrinsic to the characteristic (shared with evaluators, encoded into the cache key).

  • computation_options (tuple[ComputationOption, ...]) – Options controlling the numerical algorithm (fitter-specific).

  • constraint_tags (frozenset[str]) – Constraint tags used for matching (e.g. {"continuous", "univariate"}).

  • description (str) – Human-readable summary of what the fitter does.

Notes

The combined options property returns all options (characteristic first, then computation) for backwards-compatible resolution.

fitter: FitterFunc
to_computation_method()[source]

Build a FitterMethod (computation method) from this descriptor.

Return type:

FitterMethod

__init__(name, target, sources, characteristic_options=(), computation_options=(), constraint_tags=<factory>, description='', fitter=None)
Parameters:
Return type:

None