How to combine tar_force() with dynamic branching?

[MxNl]

I need to combine tar_force() with dynamic branching. Currently, I don't see a pattern argument in tar_force(). How can I do this? Thanks a lot as usual for you great support!

As a workaround I thought of using a helper target like

tar_force(
   target_name, 
   Sys.time(),
   force = !dbExistsTable(...)
),
tar_target(
   create_table_in_db
)

to create a table in a database if it doesnt already exist.

[wlandau]

I am avoiding built-in dynamic branching in tar_force(), tar_change(), etc. because there is potential for confusion. Would you elaborate on your use case?

[wlandau]

Glad it works for you. I just added a new tar_older() function to targets, which should simplify this to something like tar_invalidate(all_of(tar_older(Sys.time - as.difftime(1, units = "weeks")))) .

[gofford]

Cues and cue-based functionality are defined statically and are not compatible with dynamic branching.

Does this mean that tar_cue_age() can be expected to work on a per-branch basis if the branches are generated statically by tar_map()? I have a scenario where this isn't happening, but want to validate the expected behaviour before I raise a separate discussion to troubleshoot.

[wlandau]

Yes, all cues are compatible with static branching.

[gofford]

So if I have something like this, where model_index is pulled in from a map:

tar_target_raw(
   name = "model_fitted",
      command = substitute(
         expr = lightgbm_forecaster(split, model_index, target, nthread, archive_path),
         env = list(
            split = quote(model_data$split),
                model_index = quote(model_index),
                target = config$model$target,
                nthread = config$nthread,
                archive_path = config$targets$folders$models
             )
         ),
         cue = tar_cue_age_raw(
            name = "model_fitted",
            age = as.difftime(7, units = 'days')
         )
     )

expectation is that each of the model_fitted_{0,1....16} will be invalidated based on their own individual age, rather than across all of the different mapped elements?

[wlandau]

Hmm... I actually take that back tar_map() modifies the command and pattern part of targets, but not tar_cue_age(). tar_cue_age() needs to be unique to each static branch because it needs the target name. But tar_eval() should work: https://books.ropensci.org/targets/static.html#metaprogramming