AbstractLikelihood

class pmrf.likelihoods.AbstractLikelihood

Bases: Module

Abstract base class for likelihood models.

A likelihood in ParamRF specifies a mapping from model predictions to a probability over observed data. It operates in “event space”, where the probabilistic event, such as frequency, is the last axis.

This works for both deterministic and probabilistic models (e.g. Gaussian processes):

• For deterministic predictions, returns the conditional distribution \(p(y_{true} \mid y_{pred})\).

• For probabilistic predictions, returns the marginal distribution \(p(y_{true})\) over \(y_{pred}\).

See pmrf.likelihoods for built-in likelihood models.

abstractmethod __call__(y_event: Array | AbstractDistribution) → AbstractDistribution

Evaluate the likelihood given model predictions.

Parameters:

y_event (jnp.ndarray | AbstractDistribution) – The model prediction or predictive distribution in event space.

Returns:

The probability distribution over the observed data.

Return type:

AbstractDistribution