Hi Metaflow, I don’t have metaflow UI set up right now, but I was wondering: 1. Is there any functionality for sorting runs by data artifacts? e.g. I want the run with the best model eval score 2. What are tools for continual model evaluation? e.g. I want to see how the previous model would have done vs. the current model on the latest set of evaluation data

Wrt 1: If you’re training on hypers within Metaflow, a common pattern is for each then join runs, pick the best on eval and run on test for a generalization measure - eg https://github.com/jacopotagliabue/recs-at-resonable-scale Wrt 2: it’s easy to build such a thing with the client API. However in my experience it may not be worth the trouble as compared to plug-in an experiment tracker tool

@worried-mechanic-36312 , thanks! What experiment tracker tool do you have in mind?

There are many excellent choices: this features metaflow with Neptune and Wandb https://github.com/jacopotagliabue, this https://github.com/jacopotagliabue/post-modern-stack includes Comet