`modeltime_calibrate()` Unable to Calibrate a Linear Regression Model (R 4.3.1) - `.pred_res` column returned

[domorev]

The latest version of tidyverse's predict function now returns .pred_res column instead of .pred. Because of that Modeltime is unable to calibrate the model and returns the following error:

+   modeltime_calibrate(new_data = testing(splits), quiet = FALSE)
Error: ℹ In argument: `.pred = ifelse(is.na(.pred), optins, .pred)`.
Caused by error:
! object '.pred' not found

This is the output from predict():

> lm_predictions <- workflow_fit_lm %>% 
+   predict(new_data = testing(splits))
Warning message:
In predict.lm(object = object$fit, newdata = new_data, type = "response") :
  prediction from rank-deficient fit; attr(*, "non-estim") has doubtful cases
> 
> # Typically, predictions are returned in a tibble with a `.pred` column
> lm_predictions
# A tibble: 56 × 1
   .pred_res
       <dbl>
 1      51.9
 2      63.8
 3      45.8
 4      25.4
 5      14.3
 6      14.3
 7      24.7
 8      29.2
 9      83.8
10      65.8
# ℹ 46 more rows
# ℹ Use `print(n = ...)` to see more rows

[mdancho84]

I have committed a fix. Please in a fresh Rstudio session run:

remotes::install_github("business-science/modeltime", dependencies = TRUE)

Then retry the TS Jumpstart codes.

[domorev]

Updated modeltime and ran it in a fresh session. Now produces a different error. Please see the screenshot attached.

[domorev]

> workflow_fit_lm |> modeltime_calibrate(new_data = testing(splits), quiet = FALSE)
Error: Can't subset columns that don't exist.
✖ Column `.pred_res` doesn't exist.

[mdancho84]

Not sure. Difficult to resolve since I can't reproduce and all tests are passing both locally and through GH actions.

[mdancho84]

Also a modeltime table should be piped into modeltime_calibrate(). But the error will probably be the same.

[mdancho84]

Probably what happens is downstream it's passed the forecast but now another error crips up.

So ultimately I need to figure out how to replicate the error or otherwise will need your help to create the fix.

[domorev]

Happy to help, Matt. By the way, I just re-ran the code under R 4.2. and it works. So, there must be something R version specific. Are you able to install R 4.3.1. at your end to try it?

[domorev]

[mdancho84]

Ok that is interesting. Let me try installing R 4.3 and see if I can replicate.

Give me a couple days since I have multiple things I'm juggling work-wise. Maybe this weekend.

For now proceed with R 4.2 or lower until I have a chance to fix.

[mdancho84]

This should be fixed now in the development version of modeltime. Install with:

remotes::install_github("business-science/modeltime", dependencies = TRUE)

[mdancho84]

Here's me re-running the TS Jumpstart with R4.3+ and the dev version of modeltime.

Note - there is still 1 warning because of the Rank-Deficient Matrix that results from the large number of features we use.

[mdancho84]

I'm closing- this fix is complete.

Will submit to CRAN in modeltime 1.2.7 soon.