Thank you very much for DVC and the new features from the version 2!
What would be your recommendation on how to queue experiments which depend on code changes?
The use case is to queue as individual experiment each change of the content of a Python file declared as a
deps of a DVC stage.
- one is on a branch,
- makes a change on a Python file,
dvc exp run --queuefor a DVC stage depending on this Python file,
- then do another change on the same Python file,
- then do
dvc exp run --queuefor the same DVC stage,
- and then do
dvc exp run --run-all.
It seems that committing each change is an anti-pattern and not very usable.
But if the change are not committed,
git shows the changed file hanging around as modified and the changes corresponding to an experiment seem then not tracked.
The idea would be to replicate what is below but when the changes are not on a