ref_fit

ref_fit(spec, data, outcomes, id = NULL, ...)

Fits one model per outcome. A failed outcome does not abort the panel: a non-numeric outcome (factor, character, logical) gets the status "unsupported_type", one with fewer than eight finite values or no spread "insufficient_variation", and an engine failure "nonconverged".

Arguments

Argument	Description
`spec`	A ref_spec.
`data`	A data frame of reference observations.
`outcomes`	Tidyselect specification or character vector of outcome columns. The covariate frame is kept wide.
`id`	Optional subject identifier column. It is carried on scores but is never a covariate.
`...`	Passed to the engine fitter.

Details

Covariates are the variables named in the spec’s location, scale, skew, and tail formulas; other columns are ignored. Each outcome is fitted on the rows that are complete for that outcome and those covariates, and a message reports how many rows were dropped.

Outcomes are fitted in parallel when the future.apply package is installed and a non-sequential future::plan() is active.

Value

An object of class ref_fit. fit$covariates holds the covariate names.

Examples

ref <- ref_simulate(80, seed = 1)
spec <- ref_spec(family = ref_gaussian(), location = ~ age + sex)
fit <- ref_fit(spec, data = ref, outcomes = "y")
predict(fit, newdata = ref[1:3, ], uncertainty = "conditional")

<ref_scores> 3 rows

# A tibble: 3 × 15
   .row   .id .outcome observed median centile       z tail_prob tail_surprisal
* <int> <int> <chr>       <dbl>  <dbl>   <dbl>   <dbl>     <dbl>          <dbl>
1     1     1 y            8.42   8.90   0.361 -0.355      0.722        0.325
2     2     2 y           11.5    9.38   0.940  1.56       0.120        2.12
3     3     3 y           10.1   10.1    0.496 -0.0106     0.992        0.00845
# ℹ 6 more variables: residual <dbl>, log_density <dbl>, support <chr>,
#   calibrated <lgl>, .in_sample <lgl>, status <chr>