Create an empty calibration set instead of erroring

[hfrick]

closes #570

This PR adds a mock calibration split which returns the (outer) analysis set as the (inner) analysis set and an empty, i.e., 0-row calibration set. This is used when the calibration split isn't possible, e.g., if there is only one row in the analysis set.

Below is an example with the Chicago data. A lookback = 0 means we only have one period in the analysis set and thus can't split further using the principles we settled on. This is the slightly less obvious case corresponding to "only one row available to split".

library(rsample)
data(Chicago, package = "modeldata")

chicago_r_set <- sliding_period(
  Chicago,
  index = "date",
  period = "month",
  lookback = 0,
  assess_stop = 1,
  step = 2
)

chicago_r_split <- get_rsplit(chicago_r_set, 1)

i_split <- inner_split(chicago_r_split, .get_split_args(chicago_r_set))
#> Warning: Cannot create calibration split; creating an empty calibration set.

# empty calibration set
assessment(i_split)
#> # A tibble: 0 × 50
#> # ℹ 50 variables: ridership <dbl>, Austin <dbl>, Quincy_Wells <dbl>,
#> #   Belmont <dbl>, Archer_35th <dbl>, Oak_Park <dbl>, Western <dbl>,
#> #   Clark_Lake <dbl>, Clinton <dbl>, Merchandise_Mart <dbl>, Irving_Park <dbl>,
#> #   Washington_Wells <dbl>, Harlem <dbl>, Monroe <dbl>, Polk <dbl>,
#> #   Ashland <dbl>, Kedzie <dbl>, Addison <dbl>, Jefferson_Park <dbl>,
#> #   Montrose <dbl>, California <dbl>, temp_min <dbl>, temp <dbl>,
#> #   temp_max <dbl>, temp_change <dbl>, dew <dbl>, humidity <dbl>, …

^{Created on 2025-07-23 with reprex v2.1.1}

[Copilot]

Pull Request Overview

This PR implements a fallback mechanism for calibration splits that creates empty calibration sets instead of throwing errors when splits are not possible. This addresses cases where there are insufficient observations to create both an analysis set and a calibration set, particularly in time series and small dataset scenarios.

Key changes include:

• Wrapping split creation logic in try/catch blocks to handle failures gracefully
• Adding a new mock_internal_calibration_split() function to create empty calibration sets
• Converting error conditions to warnings with informative messages
• Comprehensive test coverage for all split types with mock split scenarios

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
R/inner_split.R	Core implementation of fallback logic across all split methods and new mock split function
tests/testthat/test-inner_split.R	Extensive test coverage for mock split creation across all split types
tests/testthat/_snaps/inner_split.md	Updated test snapshots showing warnings instead of errors
R/misc.R	Added class parameter to make_splits.data.frame method
man/make_splits.Rd	Updated documentation for new class parameter
inst/calibration-sets.Rmd	Documentation updates reflecting new behavior
NEWS.md	Added PR reference to changelog
DESCRIPTION	Version bump

Comments suppressed due to low confidence (1)

tests/testthat/test-inner_split.R:1205

• [nitpick] The test name is too generic. Consider a more descriptive name like 'mock_internal_calibration_split creates proper mock split structure' to clearly indicate what functionality is being tested.

test_that("can create a mock split", {