Darn. I literally just tried these out and liked it okay. Some random UX issues but nothing deal breaking.
- No setup Jupyter notebooks
- Collaborative notebook editing
- Publish notebook into AppBooks (parametrized notebooks) that allow you to deploy the resulting model, track experiments, etc. This also enables people unfamiliar with Notebooks to use it as an application without being overwhelmed, and tweak models.
- Schedule notebooks and see their output live, even if you close the browser. Your notebooks are run on a much beefier machine than when you're in interactive mode. You won't have the problem of a training job where you don't get the result because of a lost connection.
- Automatic model detection and saving, parameter/metrics tracking
- One click model deployment
- Multiple notebook checkpoints (revert back and forth to previous saved versions)
Caveat emptor: as you may notice, we're not focusing on embellishing notebooks as we believe, from what we have experienced, that the fact machine learning projects take a long time is not for lack of beautiful stylesheets. We'll tackle these problems at a later time.
- [0]: https://iko.ai
- [1]: https://www.reddit.com/r/learnmachinelearning/comments/je0pm...