sgtbrms

Extensions for brms to fit SGT model

This is a proof-of-concept extension and wrapper code to fit Skew generalised T familiy regression model. It shall work will brms and require Rstan with Stan >= 2.30.

Installation

remotes::install_github('trinhdhk/sgtbrms')

Use

An example model of how it works.

x <- rnorm(1000, 0, 1)
y <- 3*x + 3*rt(1000,3,0)
model <- brm_sgt(y ~ x, family=sgt(), prior=c(sgt_default_prior()), data=data.frame(x=x, y=y), chains=1, iter=1000)
# or equivalently
# model <- brm(y ~ x, family=sgt(), prior=c(sgt_default_prior()), stanvar=expose_sgt_stanvar(), data=data.frame(x=x, y=y), chains=1, iter=1000) 

I provide wrappers for make_stancode, make_standata and brm with suffix _sgt. These are the same calls but with sgt_default_prior() and expose_sgt_stanvar() added. You can make your own by calling:

X_sgt <- sgtbrms:::.__brm_wrap__('X')

where X is a function in brms. The special case Skew T family where p = 2 and other subsets are also provided. See help('sgt').

Change log

• 0.2: [NEW] Symmetric SGT were added. [IMPROVED] Change the parametrisation of constrained_sgt for better numerical stability.
• 0.1.2: [NEW] Constrained SGT and constrained Skew T were added.
• 0.1: [NEW] Initial PoC commit

All code is based on the implementation in the R package sgt by Carter Davis: https://cran.r-project.org/package=sgt.

Trinh Dong, 2023-2024.