Wondering if we can confirm that the version of Python being used to run our submissions matches the version being used on the VMs.
I am seeing a bug generated during my submission that I cannot seem to reproduce in my workspace. It seems to be coming from loading the saved model, which should be the same if the versions of python are the same.
We are using python 3.8.6 during evaluations. I think the issue might be related to a mismatch in the sklearn versions. Can you try pinning the version of sklearn when you install it? It should be something like
@nlevin You can specify any version of the package. The pip install commands needs to have -U flag so that the existing installation is upgraded/downgraded accordingly.