Time Series Cross Validation Warning (tidymodels, fit_resamples)

Question

I am trying to use tidymodels to run time series cross validation and have been following the Tidy Modeling with R chapters (https://www.tmwr.org/resampling#rolling).

Any help is appreciated thank you!

When I run the following code, I get "Warning message: More than one set of outcomes were used when tuning. This should never happen. Review how the outcome is specified in your model."

The resamples$.metrics and resamples$.predictions are empty except for the first [[1]].

library(tidyverse)
library(tidymodels)

df <- data.frame(
  year_month   =  seq(as.Date("2020-01-01"), as.Date("2024-03-01"), "month"),
  value  =  sample(1:100, length(seq(as.Date("2020-01-01"), as.Date("2024-03-01"), "month")))
)

splits <- sliding_window(df, lookback = Inf, skip=23, assess_start = 1, assess_stop = 1, complete = T)

model_spec <- arima_reg() %>%
  set_engine("auto_arima")

resamples <- fit_resamples(
  object       = model_spec,
  preprocessor = recipe(value ~ year_month, splits$splits[[1]]),
  resamples    = splits,
  control      = control_resamples(save_pred = TRUE)
)

collect_metrics(resamples)

resamples %>%
  select(id, .predictions) %>%
  unnest(.predictions)

Answer 1

Duplicate of this post on Posit Community. :) Quoting that answer:

I'm able to replicate the issue you're seeing with the CRAN version of tune. However, if I update my tune version to the current dev version (currently 1.1.2.9021) by running devtools::install_github("tidymodels/tune") and then restarting R, the warning about >1 outcomes goes away. That tune version will be on CRAN by the end of the month.

That said, some issues remain with this approach (note the number of rows in assessment(splits$splits[[1]]) and the message in show_notes(resamples)), but I'll let you troubleshoot from there.