Блог*

Forwarded from AlexTCH

https://github.com/marimo-team/marimo

An alternative to good (not really) 'ol Jupyter Notebooks.

Looks promising. First of all, they solve the biggest Jupyter problem: stateful execution and thus dependence on the order of cell evaluation. The problem is so large they trained ML models to predict the correct evaluation order. I kid you not.

The solution is pretty common and straightforward: upon loading a file, build the Dataflow Graph, then re-run all the transitive dependencies upon a cell change. This also gives you the "reactive programming for free".

Reactive updates also give you a reactive UI (almost for free). Demos look pretty nice and useful:
https://marimo.io/p/@marimo/embedding-visualizer

And while we're at it, yeah they have a Web playground, and can convert notebooks into Web pages (also slides) out-of-the-box. Moreover, they can produce dynamic Web pages without a server employing WebAssembly. I haven't tried it, but if it works well, that's immensely useful.

Among other things, they store notebooks as valid Python files, which provides integration with Git and many other tools for free. They also integrate with package managers, especially uv, and support per-project virtual environments out-of-the-box.

Additionally, they support DuckDB for running SQL queries, including on DataFrames. And implement nice UI to view, filter and summarize the results.

Overall, feels like a cool and handy, batteries included tool, alleviating a lot of Jupyter Notebooks quirks and letting you get your crap done quicker.

GitHub

GitHub - marimo-team/marimo: Transform data, train models, and run SQL with marimo — feels like a next-gen reactive notebook, stored…

Transform data, train models, and run SQL with marimo — feels like a next-gen reactive notebook, stored as Git-friendly reproducible Python. Deploy as scripts, pipelines, endpoints, and apps. All f...

👍2❤1💯1

www.tgoop.com/dereference_pointer_there/9951

801 viewsAug 9 at 13:26

https://github.com/marimo-team/marimo

An alternative to good (not really) 'ol Jupyter Notebooks.

Looks promising. First of all, they solve the biggest Jupyter problem: stateful execution and thus dependence on the order of cell evaluation. The problem is so large they trained ML models to predict the correct evaluation order. I kid you not.

The solution is pretty common and straightforward: upon loading a file, build the Dataflow Graph, then re-run all the transitive dependencies upon a cell change. This also gives you the "reactive programming for free".

Reactive updates also give you a reactive UI (almost for free). Demos look pretty nice and useful:
https://marimo.io/p/@marimo/embedding-visualizer

And while we're at it, yeah they have a Web playground, and can convert notebooks into Web pages (also slides) out-of-the-box. Moreover, they can produce dynamic Web pages without a server employing WebAssembly. I haven't tried it, but if it works well, that's immensely useful.

Among other things, they store notebooks as valid Python files, which provides integration with Git and many other tools for free. They also integrate with package managers, especially uv, and support per-project virtual environments out-of-the-box.

Additionally, they support DuckDB for running SQL queries, including on DataFrames. And implement nice UI to view, filter and summarize the results.

Overall, feels like a cool and handy, batteries included tool, alleviating a lot of Jupyter Notebooks quirks and letting you get your crap done quicker.

BY Блог*