Skip to contents

Set up priors with brms's quoting spelling

Source: R/priors.R
prior.Rd

prior() is set_prior() with the distribution given UNQUOTED, as brms's prior() takes it: prior(normal(5000, 1000), nlpar = "ult") is set_prior("normal(5000, 1000)", nlpar = "ult"). Every argument is deparsed rather than evaluated, so class = b and class = "b" mean the same thing, and a variable holding a distribution is deparsed to its NAME rather than its value: use prior_string() to build a prior from strings computed at run time.

Usage

prior(prior, ...)

prior_(prior, ...)

prior_string(prior, ...)

Arguments

prior

Distribution string, e.g. "normal(0, 5)", or a prior_normal()/prior_t()/prior_lkj() object, or "" for bounds only.

...

Any of set_prior()'s remaining arguments: class, coef, group, resp, dpar, nlpar, lb, ub.

Value

A frmtmb_priorlist.

Details

prior_() takes one-sided formulas, calls, names or constants (prior_(~normal(0, 10), class = ~b)) and prior_string() takes plain strings; both exist so that priors can be built programmatically, and both are brms's.

A frmtmb prior and a brms prior are different objects, and with brms attached after frmtmb its prior() masks this one. Nothing breaks: frm() and frmtmb.sample::frm_sample() accept a brmsprior object and translate its rows, so c(prior(...), prior(...)) copied out of a brms script works whichever prior() was in scope.

Examples

# the brms nonlinear vignette's spelling
prior(normal(5000, 1000), nlpar = "ult")
#> b_ult ~ normal(5000, 1000)

# combine with c() or `+`, as with set_prior()
c(prior(normal(1, 2), nlpar = "omega"),
  prior(normal(45, 10), nlpar = "theta"))
#>           prior class coef group resp dpar nlpar   lb   ub source
#>    normal(1, 2)     b                      omega <NA> <NA>   user
#>  normal(45, 10)     b                      theta <NA> <NA>   user

# the programmatic spellings
prior_(~normal(0, 10), class = ~b)
#> b ~ normal(0, 10)
prior_string(paste0("normal(0, ", 2 * 5, ")"), class = "b")
#> b ~ normal(0, 10)