Closed simonpcouch closed 2 months ago
Related-ish to #118.
This is likely a matter of passing required_pkgs(workflow)
to .options.future
.
This may actually be a model registration issue--see https://github.com/tidymodels/parsnip/issues/1128.
Based on that reprex in parsnip, I'm going to go ahead and close—we shouldn't need to pass any future options there, instead relying on set_dependency()
calls in the model registration. If we run into this again, check that first!
When making patches to fix runs on
general-meta
, I noticed that multisession workers required that needed parsnip extension packages be loaded viaoptions.future
inside offit_members()
:https://github.com/tidymodels/stacks/blob/f2e33419b5712ae7df3ca4d689c652188dce4b59/R/fit_members.R#L157
I wonder if we could find some way to restore the current parsnip model environment
parsnip::get_model_env()
in each worker. I also wonder how/why we don't tend to see this with normal tune runs.