-
Notifications
You must be signed in to change notification settings - Fork 44
Proper .config values with parallel_over = "everything"
#1049
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
Workers are failing to start on macos-latest with
No idea why since that is what I'm using and it worked previously. Let's ignore it for now and I will restart that again in a bit. |
🎶 oops I did it again, I used the wrong target branch 🎶 |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good! About those macOS skips: didn't you recently-ish do something about conditioning a skip on the date? I think it'd be good to automatically unskip these tests soon so that we don't forget.
* switching out functions * downcasting the resamples too early * temp skips for fairness metrics * add an outcome_names method for workflow_variables preprocessors * side-quest for temp solution for fit_resamples() * update shortcut * remove namespacing from testing * case weight helper * re-enable case weights plus additional tests * check grid when setting parallel_over * note constructor * logic for adding .config in different cases * no extra space for consistency * duplicate comment * update how outcome names are returned * OMG ITS HAPPENING * more informative message * changes for #1024 * note on test case * Make `fit_resamples()` go through the nested loop (#1025) * reorganize the reference index * bring back `control_grid()` and separate out `control_last_fit()` * do loop, even without tuning parameters * config is expected to be non-empty i.e. not a 0-row tibble * Closes #1026 * remove old code --------- Co-authored-by: Max Kuhn <[email protected]> * put add_configs() back in * make sure workflow is set as trained * make sure workflow is set as trained (and do it right this time) * probably will remove later * Fix issues with melodie logging (#1029) * add in missing catalog_log() calls * add missing spacing * x$note -> note * handle multiple erros in catalog_log() * fix logging testing infrastructure * update catcher_melodie() * capture warnings and errors together * fix overcounting of errors * skip for now * more skips * namespacing * update snapshot * add skip for when we have new RNG system * add skip for when we have new RNG system * updated tests * update expected column names --------- Co-authored-by: topepo <[email protected]> * updated helpers for parallel processing (#1028) * updated helpers for parallel processing * add mirai to suggests * update skip version number * update news * Apply suggestions from code review Co-authored-by: Simon P. Couch <[email protected]> Co-authored-by: Charlie Gao <[email protected]> * Apply suggestions from code review Co-authored-by: Charlie Gao <[email protected]> * updated ?parallelism * now look for installation * Update R/parallel_new.R Co-authored-by: Simon P. Couch <[email protected]> * Update R/parallel_new.R Co-authored-by: Simon P. Couch <[email protected]> * Update R/parallel_new.R Co-authored-by: Simon P. Couch <[email protected]> * refactoring --------- Co-authored-by: Simon P. Couch <[email protected]> Co-authored-by: Charlie Gao <[email protected]> * Enable mirai for parallel processing (#1032) * move functions to parallel.R * add new looping code * looping call tests * bring mirai into parallel processing * update snapshot * refactor loop_call * add missing snapshots * Update R/parallel_new.R Co-authored-by: Charlie Gao <[email protected]> * add environment back --------- Co-authored-by: Charlie Gao <[email protected]> * Parallel seeds (#1033) * updates to enable seeds in the worker processes * reset seed * more tests * typo * update news file * add more tests for rng stream * Apply suggestions from code review Co-authored-by: Simon P. Couch <[email protected]> * Apply suggestions from code review --------- Co-authored-by: Simon P. Couch <[email protected]> * small cleanup * int_pctl() updates (#1037) * refactored processing for percentile intervals * add replicate values * added another note to news * add version number * Update NEWS.md Co-authored-by: Simon P. Couch <[email protected]> --------- Co-authored-by: Simon P. Couch <[email protected]> * remove seed column * many more logging related fixes (#1035) Co-authored-by: topepo <[email protected]> close #1031 * Even more tune bug hunting (#1041) * some some skips * remocve two skips * tune_env -> melodie_env * add progress started element to melodie_env * No extra rng seed alterations for last_fit() (#1043) * changes for #1038 * typo * Update R/parallel_new.R Co-authored-by: Simon P. Couch <[email protected]> * reworked text --------- Co-authored-by: Simon P. Couch <[email protected]> * Logging bayes (#1046) * Fix spacing logging (#1052) * use correct padding * update snapshots * Proper .config values with `parallel_over = "everything"` (#1049) * changes for #1045 * update tests * temporary skips due to GHA issue * change to _time_ and OS-based skip * temporarily add our working branch to use all tests (#1053) * cleaning up the tune/melodie code (#1055) Co-authored-by: Emil Hvitfeldt <[email protected]> * Refine new logging (#1056) * move over melodie logging to tune logging (#1058) * version bump and news * use new accessor function `calibration()` (#1060) * remove leftover logic * `inner_split()` -> `internal_calibration_split()` (#1061) * `inner_split()` -> `internal_calibration_split()` * update remote --------- Co-authored-by: Hannah Frick <[email protected]> Co-authored-by: Emil Hvitfeldt <[email protected]> Co-authored-by: Simon P. Couch <[email protected]> Co-authored-by: Charlie Gao <[email protected]> Co-authored-by: Emil Hvitfeldt <[email protected]>
Closes #1045
The issue was: we were creating the
.config
values inside the tuning loops. When usingparallel_over = “resamples"
, the loop had the entire grid. Withparallel_over = “everything"
, only a subset of grid candidates were available within the loop.To fix this, we’re going to pre-compute the
.config
values prior to looping. Since those don’t change, they are added to thestatic
object. Unfortunately, many of the tests for internal functions (that assumestatic
is already made) have to be updated, too, hence the large number of altered files.