I'm talking about tools similar to Comet ML or MLFlow. I'm also interested in hearing about your experiences with them if possible.
Ask HN: What are some good Machine Learning Experiment tracking platforms?
1–4 of 4 posts
Re: Ask HN: What are some good Machine Learning Experiment tracking platforms?
#2I 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)
Re: Ask HN: What are some good Machine Learning Experiment tracking platforms?
#3This is probably not what you're looking for, but I'll describe my setup. I do CT research, and would basically store information about how an image was generated in the filename and log files. This didn't scale well, so I developed https://github.com/stilley2/filesdb which is a wrapper on Python's sqlite library. It allows me keep track of the parameters used to generate each image and easily search/locate them.
Re: Ask HN: What are some good Machine Learning Experiment tracking platforms?
#4Some tracking and training tools mentioned here:
https://medium.com/@iman.mirzadeh/4-helpful-websites-for-mac...