mixle.inference.priors moduleΒΆ

Prior specifications for declaration-backed MAP fitting.

The classes in this module are lightweight, serializable descriptions of common priors. They intentionally do not depend on NumPy, Torch, or a concrete compute engine; fitting code turns them into backend tensors at objective time.

class AlignedPrior(prior, alignment_receipt)[source]

Bases: object

Apply one prior through an explicitly acknowledged alignment mode.

Parameters:

  • prior (Any)

  • alignment_receipt (PriorAlignmentReceipt)

as_dict()[source]

Return the wrapped prior and its durable alignment receipt.

Return type:

dict[str, Any]

class BetaPrior(alpha, beta, parameter=None, improper_receipt=None)[source]

Bases: object

Beta prior for unit-interval parameters.

Parameters:

  • alpha (float)

  • beta (float)

  • parameter (str | None)

  • improper_receipt (ImproperPriorReceipt | None)

as_dict()[source]

Return the normalized Beta prior payload.

Return type:

dict

class ConditionalPrior(conditions, default=None, given=None, alignment_receipt=None)[source]

Bases: object

Per-key, default, and given priors for a ConditionalDistribution.

Parameters:

  • conditions (Mapping[Any, Any])

  • default (Any | None)

  • given (Any | None)

  • alignment_receipt (PriorAlignmentReceipt | None)

as_dict()[source]

Return keyed/default/given priors as a plain payload.

Return type:

dict

class CompositePrior(children, alignment_receipt=None)[source]

Bases: object

Child priors for a CompositeDistribution.

Parameters:

  • children (Sequence[Any])

  • alignment_receipt (PriorAlignmentReceipt | None)

as_dict()[source]

Return child priors as a plain composite-prior payload.

Return type:

dict

class DirichletPrior(alpha, improper_receipt=None)[source]

Bases: object

Dirichlet prior for simplex-valued parameters.

Parameters:

  • alpha (Any)

  • improper_receipt (ImproperPriorReceipt | None)

as_dict()[source]

Return the normalized Dirichlet prior payload.

Return type:

dict

class GammaPrior(shape, rate, parameter=None, improper_receipt=None)[source]

Bases: object

Gamma prior for positive scalar parameters or ordered-bound deltas.

Parameters:

  • shape (float)

  • rate (float)

  • parameter (str | None)

  • improper_receipt (ImproperPriorReceipt | None)

as_dict()[source]

Return the normalized Gamma prior payload.

Return type:

dict

class ImproperPriorReceipt(rationale)[source]

Bases: object

Explicit acknowledgement required to construct an improper limiting prior.

Improper priors are not probability distributions and can yield an improper posterior. The non-empty rationale makes that deliberate modeling choice durable in serialized prior payloads instead of letting a zero hyperparameter silently change the statistical contract.

Parameters:

rationale (str)

as_dict()[source]

Return the durable improper-limit acknowledgement.

Return type:

dict[str, Any]

class MarkovChainPrior(initial=None, transitions=None, length=None, alignment_receipt=None)[source]

Bases: object

Initial, transition-row, and length priors for MarkovChainDistribution.

Parameters:

  • initial (Any | None)

  • transitions (Mapping[Any, Any] | None)

  • length (Any | None)

  • alignment_receipt (PriorAlignmentReceipt | None)

as_dict()[source]

Return initial, transition, and length priors as a plain payload.

Return type:

dict

class MixturePrior(components=(), weights=None, alignment_receipt=None)[source]

Bases: object

Component and weight priors for a MixtureDistribution.

Parameters:

  • components (Sequence[Any])

  • weights (Any | None)

  • alignment_receipt (PriorAlignmentReceipt | None)

as_dict()[source]

Return component and weight priors as a plain payload.

Return type:

dict

class NormalGammaPrior(mu0=0.0, kappa=1e-06, alpha=2.0, beta=1.0, improper_receipt=None)[source]

Bases: object

Normal-Gamma prior for Gaussian mu and precision tau.

The density is proportional to tau ** (alpha - 1) exp(-beta tau) sqrt(tau) exp(-0.5 * kappa * tau * (mu - mu0) ** 2). Normalizing constants are omitted because MAP fitting only needs objective differences.

Parameters:

  • mu0 (float)

  • kappa (float)

  • alpha (float)

  • beta (float)

  • improper_receipt (ImproperPriorReceipt | None)

as_dict()[source]

Return the normalized prior payload consumed by MAP fitters.

Return type:

dict

to_distribution()[source]

Convert a proper specification to the stats-layer conjugate distribution.

Return type:

Any

class OptionalPrior(observed=None, missing=None)[source]

Bases: object

Observed-child and missing-probability priors for OptionalDistribution.

Parameters:

  • observed (Any | None)

  • missing (Any | None)

as_dict()[source]

Return observed-child and missingness priors as a plain payload.

Return type:

dict

class PriorAlignmentReceipt(mode, rationale)[source]

Bases: object

Durable acknowledgement that a structural prior is partial or broadcast.

Parameters:
as_dict()[source]

Return the serialized alignment acknowledgement.

Return type:

dict[str, str]

class RecordPrior(fields, alignment_receipt=None)[source]

Bases: object

Field priors for a named RecordDistribution.

Parameters:

  • fields (Mapping[Any, Any])

  • alignment_receipt (PriorAlignmentReceipt | None)

as_dict()[source]

Return field priors as a plain record-prior payload.

Return type:

dict

as_prior_dict(prior)[source]

Return a plain-Python representation of a prior specification.

Parameters:

prior (Any)

Return type:

Any

beta(alpha, beta, parameter=None, improper_receipt=None)[source]

Create a Beta prior for a unit-interval parameter.

Parameters:

  • alpha (float)

  • beta (float)

  • parameter (str | None)

  • improper_receipt (ImproperPriorReceipt | None)

Return type:

BetaPrior

broadcast(prior, rationale)[source]

Wrap a prior for intentional broadcast across a model structure.

Parameters:
Return type:

AlignedPrior

broadcast_alignment(rationale)[source]

Acknowledge that one prior intentionally applies to every structural position.

Parameters:

rationale (str)

Return type:

PriorAlignmentReceipt

conditional(conditions, default=None, given=None, alignment_receipt=None)[source]

Create a Conditional prior over keyed/default/given child priors.

Parameters:

  • conditions (Mapping[Any, Any])

  • default (Any | None)

  • given (Any | None)

  • alignment_receipt (PriorAlignmentReceipt | None)

Return type:

ConditionalPrior

composite(children, alignment_receipt=None)[source]

Create a Composite prior from child prior specifications.

Parameters:

  • children (Sequence[Any])

  • alignment_receipt (PriorAlignmentReceipt | None)

Return type:

CompositePrior

dirichlet(alpha, improper_receipt=None)[source]

Create a Dirichlet prior for simplex-valued parameters.

Parameters:

  • alpha (Any)

  • improper_receipt (ImproperPriorReceipt | None)

Return type:

DirichletPrior

gamma(shape, rate, parameter=None, improper_receipt=None)[source]

Create a Gamma prior for a positive parameter.

Parameters:

  • shape (float)

  • rate (float)

  • parameter (str | None)

  • improper_receipt (ImproperPriorReceipt | None)

Return type:

GammaPrior

improper(rationale)[source]

Create the explicit receipt required for a supported improper limiting prior.

Parameters:

rationale (str)

Return type:

ImproperPriorReceipt

markov_chain(initial=None, transitions=None, length=None, alignment_receipt=None)[source]

Create a Markov-chain prior over initial, transition, and length terms.

Parameters:

  • initial (Any | None)

  • transitions (Mapping[Any, Any] | None)

  • length (Any | None)

  • alignment_receipt (PriorAlignmentReceipt | None)

Return type:

MarkovChainPrior

mixture(components=(), weights=None, alignment_receipt=None)[source]

Create a Mixture prior over component and weight priors.

Parameters:

  • components (Sequence[Any])

  • weights (Any | None)

  • alignment_receipt (PriorAlignmentReceipt | None)

Return type:

MixturePrior

normal_gamma(mu0=0.0, kappa=1.0e-6, alpha=2.0, beta=1.0, improper_receipt=None)[source]

Create a Normal-Gamma prior for Gaussian mean/precision parameters.

Parameters:

  • mu0 (float)

  • kappa (float)

  • alpha (float)

  • beta (float)

  • improper_receipt (ImproperPriorReceipt | None)

Return type:

NormalGammaPrior

optional(observed=None, missing=None)[source]

Create an Optional prior over observed child and missingness terms.

Parameters:

  • observed (Any | None)

  • missing (Any | None)

Return type:

OptionalPrior

partial_alignment(rationale)[source]

Acknowledge that unspecified structural positions intentionally have no prior.

Parameters:

rationale (str)

Return type:

PriorAlignmentReceipt

record(fields, alignment_receipt=None)[source]

Create a Record prior from field-name prior specifications.

Parameters:

  • fields (Mapping[Any, Any])

  • alignment_receipt (PriorAlignmentReceipt | None)

Return type:

RecordPrior