Open JacksonBurns opened 1 month ago
I think this is fixed in #942. In that PR, I added task_weights = torch.ones(n_tasks) if args.task_weights is None else args.task_weights
to avoid passing None
into the generation of the loss_function
object.
That looks like it should work. Once we release 2.0.4, let's link that PR to this issue and have it automatically close. Leaving this up until then so that people know it is fixed on main
.
Describe the bug When specifying
--loss-function mse
via the CLI, aTypeError
is raised.Example(s) This code works:
but this does not:
Expected behavior Explicitly passing the default should work the same as leaving Chemprop to determine the default.
Error Stack Trace
Screenshots N/A
Environment
conda list
orpip list
Checklist
conda list
orpip list
shows the packages listed in thepyproject.toml
pytest -v
reports no errorsDidn't this check, but I just installed from PyPI and I have no local changes.
Additional context You can download the dataset I am running this on if you want to reproduce the exact error locally from this Zenodo: https://zenodo.org/records/10892219