Meta issue for survival analysis in tidymodels

[topepo]

Functionality/documentation

• (PR available) throw warning when eval_time isn't needed tune#704
• tune: autoplot() should warn when used with non-needed eval_time tune#706
• tune: Avoid warning from geom_line() bubbling up in autoplot()? tune#755
• tune: augment.tune_results() doesn't always return tibbles tune#759
• tune: augment() for censored tune_results objects doesn't let you pass eval_time to select_best() tune#762
• tune: collect_metrics() doesn't return .weight_time and .pred_censored tune#764
• yardstick: Properly document ... argument in dynamic survival metrics yardstick#427
• censored: multi_predict(type = "survival") is missing penalty column for single penalty value censored#267
• censored: Make sure all models can handle out of domain survival probability predictions  censored#10
• censored: Missing values in prediction with tree-based models censored#156
• censored: Encapsulate dealing with "all missings" into a function censored#222
• parsnip: Deprecate new_data arg for reverse_km censoring model parsnip#965
• parsnip: check dots are empty in predict.censoring_model_reverse_km() parsnip#1029
• yardstick: Integrated survival metrics should warn when used with a single time point in eval_time yardstick#460

To discuss

• Enable data-related args to survival::coxph() censored#147

Testing

• sometimes censoring weight columns have names parsnip#1023
• workflows: preprocessor initial survival workflows tests #104
• workflows: fit interfaces Test fit interfaces called from workflow for survival models #111
• workflows: Test augment.workflow() for survival models #110
• yardstick: Test that survival metrics can handle case weights #112
• tune: no tuning aka fit_resamples() Test fit_resamples() for survival models #113
• tune: last_fit() Test last_fit() for survival models #114
• tune: fit_best() Test fit_best() for survival models #115
• tune: case weights Test tuning of survival models with case weights #116
• tune: collect_metrics() Test collect_metrics() for survival models #117
• tune: show_best(), select_best() Test show_best(), select_best(), and select_by_*() for survival models #118
• tune: augment() Test augment() for tune_results, resample_results, and last_fit objects with survival models #132
• workflowsets: workflow_set() and workflow_map() Test workflow_set() and workflow_map() for survival models #119
• funetune: tune_race_*() name the evaluation time for survival models finetune#81
• Test fit_resamples() with eval_time when it's not needed #146

[hfrick]

Notes on aspects to keep in mind / good things to test:

• eval_time: single and multiple time points, unordered, non-standard (negative, infinite)
• prediction types: "time" and "survival"
• metrics: static, dynamic, integrated, mixed
• models/engines: proportional_hazards(engine = "glmnet") plus some other model
• stratification: proportional_hazards() with engines "glmnet" and "survival" do stratification
• response: recommendation is a pre-made Surv object, the alternative is on-the-fly generation a la Surv(time, status)

[hfrick]

closing this in favor of tracking properly on the board