Hi all, I have been working on a failure prediction model for the diskprediction_local module and I want to apply state-of-the-art classification/regression models for this task using packages like rgf-python, tensorflow, pytorch, etc. However, most of these packages are not available as rpm/deb packages, so if a model is built using one of these then there would be dependency issues <https://github.com/ceph/ceph/pull/29437#discussion_r309835840> with using it in a ceph module. Is there a way we can work around this? Right now I can only use the scikit-learn package, which is quite limiting if we want to build and use state-of-the-art models. Warm regards -- Karan Chauhan He/Him/His Software Engineer - AI Center Of Excellence - Office Of The CTO Red Hat <https://www.redhat.com/> kachauha@redhat.com <https://www.redhat.com/>