Skip to contents

The canonical form every named-family prior is stored in, under prior$params$marginals. lower/upper are the effective bounds, that is the family's own support intersected with whatever truncation was asked for, and log_norm is the log of the probability mass left inside them. Holding the truncated mass here rather than recomputing it per call is what keeps the density proper without paying for a CDF evaluation on every row.

Usage

new_marginal(family, args, lower = -Inf, upper = Inf, label = "the parameter")

Arguments

family

Family name, a key of prior_family().

args

Named list of scalar distribution parameters.

lower, upper

Truncation bounds, before intersecting with the family's own support.

label

How to refer to this parameter in an error message.

Value

A marginal specification.