Batched replication

[ropensci]

Batched replication

[wlandau]

Under the hood, we could create 2 targets instead of 3. This will make interaction with #13 less burdensome. We could express tar_rep_tbl(data_continuous, simulate_data_continuous(), batches = 2, reps_per_batch = 4) as:

list(
  tar_target(
    data_continuous_batch,
    seq_len(2),
    deployment = "local"
  ),
  tar_target(
    data_continuous,
    tarchetypes::tar_rep_df_run(simulate_data_continuous(), 4),
    pattern = map(index_batch),
    format = "fst_tbl"
  )
)

where

tar_rep_df_run <- function(command, reps) {
  expr <- substitute(command)
  envir <- parent.frame()
  do.call(vctrs::vec_c, tar_rep_run(expr, envir, reps))
}

tar_rep_list_run <- function(command, reps) {
  expr <- substitute(command)
  envir <- parent.frame()
  tar_rep_run(expr, envir, reps)
}

tar_rep_run <- function(expr, envir, reps) {
  lapply(seq_len(reps), tar_rep, expr = expr, envir = envir)
}

tar_rep <- function(expr, envir, rep) {
  out <- eval(expr, envir = envir)
  out$rep <- rep
  out
}

[wlandau]

If we infer the iteration type for the reps within batch from the iteration type of the target spec, then I think we can actually simplify this down into tar_rep() and tar_rep_raw().

[wlandau]

Let's call it tar_batch() instead of tar_rep().

[wlandau]

Nope: tar_rep() is better because it emphasizes that the command stays identical across batches.