Lets say I want to use the method very_nice_vi that is defined in R package bayesbench_vi_methods. Currently I would first need to run install.packages("bayesbench_vi_methods"). It would be good if this was not necessary and instead could be done automatically when I create a job that uses very_nice_vi.
pre-commit has a similar situation where it does the installation automatically. In the posteriordb repo we use the pre-commit hook trailing-whitespace. This particular hook is implemented in a python package. When you call pre-commit run for the first time the package gets automatically installed in a sandbox (so there's no chance it will conflict with globally installed packages) and on further runs the sandbox is reused. I think a similar approach might be viable for bayesbench also.
Lets say I want to use the method
very_nice_vi
that is defined in R packagebayesbench_vi_methods
. Currently I would first need to runinstall.packages("bayesbench_vi_methods")
. It would be good if this was not necessary and instead could be done automatically when I create a job that usesvery_nice_vi
.pre-commit
has a similar situation where it does the installation automatically. In theposteriordb
repo we use thepre-commit
hooktrailing-whitespace
. This particular hook is implemented in a python package. When you callpre-commit run
for the first time the package gets automatically installed in a sandbox (so there's no chance it will conflict with globally installed packages) and on further runs the sandbox is reused. I think a similar approach might be viable forbayesbench
also.