prophys.engine

Compiled models, fitting, operational surrogates and diagnostics.

CompiledModel (from ProbabilityModel.compile() or compile_model()) is the evaluation interface of a model: log densities, samples, expectations, probabilities, quantiles and tail risk of any attribute. Around it:

• fitting: calibrate(), finetune(), fit_distribution(), negative_log_likelihood(), and design optimization with optimize()

• fit checks: finite_difference_check(), check_identifiability()

• operational surrogates: amortize() and break_even()

• Monte Carlo error: assess(), the *_error helpers, required_samples() and confidence_interval()

• parameter uncertainty: parameter_uncertainty(), laplace_covariance(), propagate(), profile_likelihood(), sample_parameters()

• sensitivity: local_sensitivity(), sobol_sensitivity()

• model scoring: information_criteria(), waic(), loo(), pointwise_log_likelihood(), compare(), calibration(), crps(), posterior_predictive_check(), report_card()

The diagnostic result classes are frozen dataclasses deriving from Report, with a readable str() and a JSON ready to_dict().

Classes

CompiledModel	A probability model compiled into an evaluation interface that works with jax.jit and jax.grad.
CalibrationResult	Result of calibrate(), finetune() or fit_distribution().
OptimizationResult	Result of a design optimization with optimize().
OperationalSurrogate	A trained surrogate of a whole model's predictive distribution, with guarded fallback.
OperationalAnswer	Answers to a batch of queries from OperationalSurrogate.answer().
Guard	Gate values per row of a batch of queries, from OperationalSurrogate.guard().
AmortizationCard	Training and acceptance record of an OperationalSurrogate.
BreakEvenEstimate	Whether amortizing a model pays off, from measured or estimated costs.
Report	Base class of every diagnostic result in prophys.engine.
EstimateWithError	A Monte Carlo estimate of a mean, probability or quantile with its sampling error.
TailRiskEstimate	Value at Risk and Conditional Value at Risk estimates with the CVaR's standard error.
ParameterUncertainty	Standard errors, covariance and correlations of fitted parameters (Laplace approximation).
DerivedUncertainty	Delta method standard errors of a quantity derived from the fitted parameters.
ProfileLikelihood	Profile log likelihood of one parameter with its likelihood ratio interval.
LocalSensitivity	Derivatives and elasticities of one model statistic with respect to the parameters.
SobolSensitivity	First order and total order Sobol indices of an attribute over its uncertain inputs.
InformationCriteria	Information criteria (AIC, AICc and BIC) of a fitted model, from information_criteria().
PredictiveScore	Expected log predictive density estimated by WAIC or PSIS LOO, from waic() or loo().
ModelComparison	Ranking of models by a predictive score, with each model's difference from the best.
CalibrationReport	Probability integral transform (PIT) calibration of a predictive distribution.
CrpsScore	Continuous ranked probability score (CRPS) of a predictive distribution, from crps().
PosteriorPredictiveCheck	Posterior predictive check of one summary statistic, from posterior_predictive_check().

Functions

compile_model	Compile a probability model.
calibrate	Fit the parameters of a compiled model to observations of one attribute.
finetune	Jointly fit a compiled model against observations of several attributes.
fit_distribution	Fit the Param leaves of a standalone distribution to data by maximum likelihood.
optimize	Minimize a scalar objective over design parameters, with inequality constraints.
amortize	Train an operational surrogate of one attribute's predictive distribution over an envelope.
break_even	Estimate whether amortizing an attribute pays off, before training anything.
finite_difference_check	Compare the automatic gradient of fn against central finite differences.
check_identifiability	Check whether a loss history is still improving at the end of a fit.
confidence_interval	Confidence interval estimate +/- z * std_error from normal theory.
assess	Estimate a statistic of a model attribute together with its Monte Carlo error.
mean_error	Sample mean of draws with its batch means standard error.
quantile_error	Sample q quantile of draws with its Maritz and Jarrett standard error.
cvar_error	Upper tail CVaR of draws at level alpha, with its influence function error.
required_samples	Number of draws needed to reach a target relative error, extrapolated from an estimate.
negative_log_likelihood	Build the negative log likelihood that finetune() minimizes.
laplace_covariance	Laplace (observed information) covariance of parameters at the minimum of a loss.
parameter_uncertainty	Compute standard errors and correlations of a fit's parameters (Laplace approximation).
propagate	Propagate parameter uncertainty into a derived quantity by the delta method.
profile_likelihood	Profile the log likelihood in one parameter and read off its likelihood ratio interval.
local_sensitivity	Compute derivatives and elasticities of a model statistic with respect to its parameters.
sobol_sensitivity	Compute variance based (Sobol) sensitivity indices of an attribute to its uncertain inputs.
sample_parameters	Draw parameter vectors from the Laplace (Gaussian) approximation of the fit.
pointwise_log_likelihood	Log likelihood of every observation under every parameter draw.
information_criteria	Compute AIC, AICc and BIC for a fit of a compiled model to observations.
waic	Score predictive accuracy with the Widely Applicable Information Criterion (WAIC).
loo	Estimate leave one out predictive accuracy by Pareto smoothed importance sampling (PSIS LOO).
compare	Rank models by predictive score, with the standard error of each difference.
calibration	Test whether the predictive distribution is calibrated against observations (PIT).
crps	Score the predictive distribution with the continuous ranked probability score (CRPS).
posterior_predictive_check	Compare a summary statistic of the data against replicate datasets simulated from the model.
report_card	Run every applicable goodness of fit diagnostic for one fitted attribute.

Constants and objects

FALLBACK_REASONS

Built-in immutable sequence.