My gripe with Observable is I can't fit into a Python-based workflow. I.e., a machine learning or gpu-training worfklow. It is a nice place to create explainers, though. I almost think of it as being like a Medium alternative.
It's different to something like Observable in that you import it into your existing Python environment (Colab, Jupyter, Airflow, etc.), build the reports from your assets, and push them up -- as opposed to having to do the analysis itself in our interface.
Observable is for JavaScript ecosystem. I wouldn't count on them implementing anything for Python.
Yes I'm a fan; just would like to be able to write JS to visualize.
- Collaborative no-setup notebooks: different images with the most popular libraries pre-installed. See each other's cursors, very useful for troubleshooting, pair programming.
- Long-running notebooks: GPU ready. You can schedule notebooks right from the notebook view so you don't context switch. You can close your browser, disconnect from the internet, or shut your laptop and still see the notebook's execution.
- Automatic params/metrics/model tracking: no boilerplate.
- One click deployment: clicking a button deploys a model and gives you a nice "REST" API endpoint you can invoke.
If you're doing anything related to ML and use notebooks, you probably had the above problems. We're building around Jupyter notebooks because in our experience, our machine learning projects were not held back for lack of slick stylesheets.
- [0]: https://iko.ai