Bundle Residual Distribution Functions

Store a named parameter vector and residual distribution functions without evaluating the functions or drawing random numbers.

Usage

residual_law(param = numeric(), dresidual = NULL, presidual = NULL,
             qresidual = NULL, rresidual = NULL, name = "custom",
             density = NULL, cdf = NULL, quantile = NULL, rng = NULL)

Arguments

param

Finite numeric vector with unique, nonempty parameter names.

dresidual, presidual, qresidual, rresidual

Residual density, cumulative distribution, quantile and random-generation functions, or NULL. Use R distribution signatures: the variate, probability or sample count is the first argument. Named parameters are matched to explicit formal arguments as in emhawkes::set_residual. Log-density and tail options are supported there.

density, cdf, quantile, rng

Compatibility aliases for dresidual, presidual, qresidual and rresidual, respectively. Supply at most one non-NULL function per pair; specifying both is an error, even when they are the same function. No deprecation warning is emitted.

name

A descriptive label, not a verified distribution-family identifier.

Details

Supply only the functions needed for your operation. Construction requires at least dresidual or rresidual. Simulation needs only rresidual; univariate likelihood and fitting need only dresidual. Multivariate likelihood and fitting additionally need presidual. infer = TRUE needs dresidual, presidual and qresidual, and does not require rresidual. Missing functions are reported when an operation requires them, before callbacks or random draws. The original positional order of the first six arguments is unchanged. Stored law objects keep their existing representation, and callback functions, environments and defaults are preserved. The constructor does not establish consistency of user-supplied functions, normalization of residual scale, existence of moments, or parameter identification.

Value

An object of class frpp_residual_law.

Examples

# Simulation needs only a residual generator.
law <- residual_law(c(rate = 1), rresidual = stats::rexp)
model <- fspec(methods::new("hspec", mu = 1, alpha = 0, beta = 1), law)
fsim(model, size = 4, seed = 73, lambda_component0 = 0)
#> FRPP path: 3 events, 1 event types.

# Density alone is enough for a univariate likelihood.
law <- residual_law(c(rate = 1), dresidual = stats::dexp)
model <- fspec(methods::new("hspec", mu = 1, alpha = 0, beta = 1), law)
stats::logLik(model, c(0, .2, .3), lambda_component0 = 0)
#> loglikelihood 
#>          -0.5