glmbfamfunc: Return family functions used during simulation and post...

In glmbayes: Bayesian Generalized Linear Models (IID Samples)

Return family functions used during simulation and post processing

Description

This function takes as input a family object and returns a set of functions that are used during simulation and summarization of models using the glmb, and rglmb functions.

Usage

glmbfamfunc(family, lik_shape = 1)

## S3 method for class 'glmbfamfunc'
print(x, ...)

Arguments

family	an object of class family
lik_shape	Known shape parameter of the Gamma likelihood; used only for the Gamma(link = "identity") branch where the regression coefficient is the Gamma rate. Ignored for all other family/link combinations. Defaults to 1 (i.e.\ exponential likelihood).
x	an object of class "glmbfamfunc" for which a printed output is desired.
...	additional optional arguments

Details

glmbfamfunc is the canonical R closure bundle for likelihood, posterior, gradient, and deviance quantities across all supported family/link combinations.

Registration requirement. A branch must exist inside glmbfamfunc for every family/link combination that is used in the package. If a combination is missing, the closures f1–f4 will never be assigned and any downstream code that accesses them will fail with "object 'f1' not found". New family/link combinations must therefore be explicitly added here before they can produce valid DIC, log-likelihood, or directional-tail output.

Currently implemented family/link combinations:

Family	Link
gaussian	identity
poisson, quasipoisson	log
poisson, quasipoisson	identity
binomial, quasibinomial	logit
binomial, quasibinomial	probit
binomial, quasibinomial	cloglog
binomial, quasibinomial	identity
Gamma	log
Gamma	identity

Any family/link not in this table will fall through all branches silently and produce the error above at the point of first use. For Gamma(link = "identity"), pass the known Gamma likelihood shape lik_shape (default 1) so that f1–f4 and f7 use the correct parameterization (coefficient = Gamma rate \beta).

Relationship to C++ simulation paths. Many simulation procedures in the package have been fully or partially migrated to *.cpp routines, which receive their own objective functions directly. For those paths glmbfamfunc may not be called at all during sampling. However, the R-side post-processing functions (logLik.glmb, summary.rglmb, directional_tail) always use famfunc\$f1, famfunc\$f4, and famfunc\$f7 respectively, so a registered branch is still required for those outputs even when the sampler itself has moved to C++.

Value

A list (class "glmbfamfunc") whose first four components are always present for every supported family and link. The names f1–f4 are stable: they mean the same roles across families (only the internal formulas change).

f1

Negative log-likelihood as a function of coefficients b (arguments typically b, y, x, optional alpha, wt).

f2

Negative log-posterior (likelihood plus Normal prior quadratic form in b with precision P and mean mu).

f3

Gradient of f2 with respect to b (same argument pattern as f2).

f4

Deviance-related quantity (twice negative log-likelihood contrast vs.\ saturated model, with a dispersion argument for quasi-families); used in DIC-style summaries.

f7

Family-specific matrix: weighted sum of outer products of predictor rows, i.e.\ a curvature / expected negative Hessian of the log-likelihood w.r.t.\ b at the supplied b (used e.g.\ by directional_tail for glmb fits).

Slots f5 and f6 are not returned: they were reserved for alternate or C++-aligned likelihood/posterior routines and remain commented out in the implementation (only f1, f2, f3, f4, and f7 are assigned in the returned list).

Examples

famfunc <- glmbfamfunc(binomial(logit))

print(famfunc)

## f1--f4 and f7 are always present for supported families; f5 and f6 are not returned.
f1 <- famfunc$f1
f2 <- famfunc$f2
f3 <- famfunc$f3
f4 <- famfunc$f4
f7 <- famfunc$f7
stopifnot(is.function(f1), is.function(f2), is.function(f3),
          is.function(f4), is.function(f7))

glmbayes documentation built on Aug. 5, 2026, 1:07 a.m.