A run's ignored files are kept under tune/<key>/<setup>/, one folder per
setup compared side by side (#119). The setup is the four tokens of link's
output vocabulary its build logged (cut.acc.grad.flow, from the grade
record), each held to link's vocabulary.csv when one is given. A build
from before link 0.70.0 logged none and is kept under
untokened/<tune run>: its folder is never renamed when its tokens are
named later (link#420); the index rows gain them instead. A re-run of a
tokened setup overwrites its folder, and the bucket keeps the previous
objects as versions.
Arguments
- record
a grade record (
knw_grade_record_read()).- vocabulary
NULL, or link's vocabulary (tokenandaxiscolumns, link'sinst/extdata/vocabulary.csv).
Examples
knw_archive_setup(data.frame(token_cut = NA, token_acc = NA, token_grad = NA, token_flow = NA,
tune_run = "20261027_0000"))
#> [1] "untokened/20261027_0000"
knw_archive_setup(data.frame(token_cut = "cut-full-top", token_acc = "acc-v1", token_grad = "grad-seg",
token_flow = "q-wet-fill", tune_run = "20261101_0000"))
#> [1] "cut-full-top.acc-v1.grad-seg.q-wet-fill"
