Update docs for fit.workflow(calibration) #292
Description
Generally, we want to be able to pass a subset of our data as the calibration set to fit.workflow() and expect workflows to handle the step of generating predictions for those observations and then use those to fit the calibrator. This also came up in #289.
The documentation for calibration currently says that the dataset needs to already contain those predictions. That's rather inconvenient when you are just trying to fit your primary model!
Lines 23 to 26 in 835ee35
Poking around in fit.action_tailor(), there is a call to generate the predictions already, so we need to update the docs.
workflows/R/post-action-tailor.R
Lines 135 to 145 in 835ee35
Reprex of it working:
library(tidymodels)
library(tailor)
set.seed(1)
dat <- sim_regression(200)
split <- initial_split(dat)
rf_spec <- rand_forest(mode = "regression", trees = 20)
tlr <- tailor() |> adjust_numeric_calibration(method = "linear")
rf_wflow <- workflow(outcome ~ ., rf_spec, tlr)
# set aside data for calibration
set.seed(2)
i_split <- inner_split(split, .get_split_args(split))
i_analysis_dat <- analysis(i_split)
cal_dat <- assessment(i_split)
manual_fit <- fit(rf_wflow, data = i_analysis_dat, calibration = cal_dat)
#> Registered S3 method overwritten by 'butcher':
#> method from
#> as.character.dev_topic genericsCreated on 2025-07-16 with reprex v2.1.1