objectives (pmrf.objectives)

Public objectives modules and their common entry points.

Modules

evaluators	Callables that evaluate a model over frequency.
losses	Loss models for optimizer fitting or generalized Bayesian inference.
problems	A callable to be "solved" (i.e. minimized or sampled).
terms	Callables that evaluate a parameter PyTree, which form the summands of a problem.

Classes

AbstractEvaluator(*[, name, metadata])	Abstract base class for callables that evaluate a model over frequency.
AbstractLoss(*[, name, metadata])	Abstract base class for frequentist loss functions.
AbstractProblem()	Abstract base class for callables to be "solved" (i.e. minimized or sampled).
AbstractTerm(*[, name, metadata])	Abstract base class for callables that evaluate a model.
BoundEvaluator(evaluator, frequency[, ...])	Binds an evaluator to the frequency sweep it should be evaluated over.
Feature(alias)	Extracts an RF feature using a string-based alias.
GibbsMarginalLogLikelihood(predictor, ...[, ...])	Computes a generalized log-posterior (the Gibbs measure) using a loss function instead of a strict generative likelihood.
Goal(feature[, operator, target, weight, ...])	Computes a design goal using a hinge-based loss evaluator.
HingeLoss([operator, weight, mask, ...])	Applies a one-sided constraint (hinge) before evaluating a base metric.
HuberLoss([delta, multioutput, name, metadata])	Huber loss metric.
LogMSELoss([multioutput, name, metadata])	Log of Mean Squared Error (RMSE) metric.
MAPELoss([multioutput, name, metadata])	Mean Absolute Percentage Error (MAPE) metric.
MSELoss([multioutput, name, metadata])	Mean Squared Error (MSE) metric.
MarginalLogLikelihood(predictor, observed, ...)	Computes the log of the probability of observed data by conditioning a likelihood function on a model prediction while marginalizing out a potential discrepancy model.
Negated(evaluator, *[, name, metadata])	Computes the negative of another evaluator.
PriorPenalized(problem)	A problem penalized by the negative log prior of its parameters, in declared space.
RMSELoss([multioutput, name, metadata])	Root Mean Squared Error (RMSE) metric.
SummedTerms(model, terms)	A problem summing a set of terms over a tree of parameters.
TargetLoss(predictor, target, loss, *[, ...])	Computes a loss between a model prediction and some target.

Functions

as_terms(objective[, frequency])	Normalize a user-supplied objective into a tuple of terms.
problem_terms(problem, name)	Returns the terms of a problem, looking through any wrappers.