Open WouterDH-UZL opened 7 months ago
Dropping in to confirm that I see this as well and that this is possibly related to #23 and HenrikBengtsson/doFuture#82. The issue in this case is that load_pkgs()
is called as load_pkgs(workflow)
--where workflow
is a local variabe--but workflow
is being passed as workflows::workflow
. We see the aforementioned error rather than object '' not found
because even though the worker isn't able to find the local variable, it can find the function exported from workflows.
Describe the bug I wasn't sure whether to post this here or in the 'tidymodels' repo, but using future.callr does not seem to work in conjunction with tune::tune_grid(). I've been using future.callr to stop grid searches and model building from consuming all GPU RAM, and it's been working great thus far. However, combining this approach with the tidymodels framework leads to issues.
Reproduce example
Using
tune_grid()
withplan(callr)
leads to an error that seems to be due to a namespacing issue:Expected behavior The function tune::tune_grid() has support for multithreading via futures (using 'doFuture'), so I expected plan(callr) to work without issue (as the other plan() calls do).
Session information