I don't have experience with this yet, but the CEO of comet.ml gave a demo of it at work this week and it's brilliant. You install 1 pip package, add 1 line of

I don't have experience with this yet, but the CEO of comet.ml gave a demo of it at work this week and it's brilliant. You install 1 pip package, add 1 line of code, and you've got a log every time you train your model (with a diff of what changed). Super-easy to use and share with your team. I'm thinking one cool use case is: easily let several team-mates try out different model hyperparameters, and you just pick the best one (easy to see in the comet.ml online chart)