Resilient seeds for batched replication

[wlandau]

Prework

• I understand and agree to this repository's code of conduct.
• I understand and agree to this repository's contributing guidelines.
• I have already submitted an issue or discussion thread to discuss my idea with the maintainer.

Related GitHub issues and pull requests

• Ref: Rep-specific seeds in tar_rep(), tar_map_rep(), etc. #111

Summary

In targets, target-specific pseudo-random number generator seeds are deterministic and depend on the target names. So far, these default seeds have applied to entire batches of replications in tarchetypes target factories like tar_rep() and tar_map_rep(). This behavior has had the undesirable consequence of changing seed assignment when the batching structure changes (i.e. when batches and reps change while the total number of replications batches * reps remains constant).

This PR assigns a special seed to each replicate. These new seeds depend on the parent target name and the total rep index. As long as batches * reps remains constant, these seeds will not change if you change the batching structure. In other words, tar_rep(name = x, command = rnorm(1), batches = 100, reps = 1, ...) now has the same output as tar_rep(name = x, command = rnorm(1), batches = 10, reps = 10, ...). Seeds are available in the output of most target factories through the "tar_seed" column.

The affected functions are:

• tar_rep()
• tar_rep2()
• tar_map_rep()
• tar_map2_count()
• tar_map2_size()
• tar_render_rep()

For tar_map2_count() and tar_map2_size(), it is also possible to generate your own seeds in command1 and use them in command2. Similarly, you can supply seeds in tar_render_rep() via the params argument and then use them in the R Markdown report. Likewise in tar_quarto_rep() with execute_params. tar_quarto_rep() is ready on tarchetypes' end for seed resilience, but Quarto itself does not currently have a flag to set seeds.