Open GenevieveBuckley opened 11 months ago
(For extra context: our HPC team have said that this is a blocker for use (a persistent service requiring execution from a privileged account is not a good fit for shared use HPC systems, and there are possible cybersecurity risks that limit our options. So it would be nice if we could work around that issue altogether)
Hi Genevieve, @patrickstock is working on making jupyter notebooks (potentially even cloud hosted on Colab/Sagemaker) for inference (WIP) and training (coming soon) that don't talk to an experiment tracking database, see here. This will result in getting rid of the mongoDB dependency entirely. We'll be happy to notify you once this is ready for trying it out.
Very cool, we'd love to try that out when it's ready
Following up on an earlier conversation with @bentaculum about removing MongoDB as a dependency.
What exactly would be involved with something like this? Eg: Is a replacement required or can it be removed entirely, what technical issues might this cause, etc. I see that pymongo is used in the train, predict, and predict_multiple scripts, but I don't have a lot of insight beyond that.