Survival analysis updates

[topepo]

Update the tune functions here to keep up with the changes in the tune.

These are pretty straightforward changes. Testing for survival models will be in extratests (after this is merged, step 6 below)

An updated itinerary of PRs

1. Tools for selecting a default evaluation time tune#768
2. Refactor choose_metric tune#777
3. Refactored functions to select single evaluation times tune#778
4. Refactored functions to evaluate/check metric and eval time arguments tune#780
5. (you are here) update finetune to use these changes
6. Unit tests that can't go into tune/finetune in the extratests repo
7. Revist some of the checking functions and messaging (e.g., Finetuning warnings for show_best() on results of fit_resamples() and other tune_results tune#781)
8. Add additional functions that return (potentially) multiple evaluation times (eg autoplot())
9. Find remaining unused functions in tune (like middle_eval_time()) to remove
10. Write an article on tidymodels.org about this.

[hfrick]

same question about the call here

[topepo]

☝️

[hfrick]

x and y are both tune_results objects, right? do they now both carry eval_time_target as an attribute that could be used here instead of providing it as an argument?

[topepo]

x has class tune_results. y has the same structure but different attributes. The extra classes have been dropped due to row filtering of the resampling object. The function restores missing attributes to y.

However... x is from tune_grid(), which has no notion of a target evaluation time. tidymodels/tune#782 defaults eval_time_target to NULL for grid tuning, resampling, and last fit objects.

That's why eval_time_target is an argument to this function. It should be non-null for the resulting racing object but the value inherited from x is NULL unless we set it.

[hfrick]

Ah, thanks! 👍

[topepo]

notes added to sources