Model and Training set version control

[patcdaniel]

I've been thinking about model and training set version control and wondering if anyone here has any insight into best practices. One use case might be for an operational classifying model periodically retrained as training sets improve once or twice a year.

I saw some versioning capability on Zenodo, which might make sense as models are published, but what about during development or collaboration? DVC might be a solution help with collaboration, but I am curious about what others are doing.