The New World of Notebook Publishing | Hacker News Reader

The New World of Notebook Publishing

writings.stephenwolfram.com

79 pointstosh29 comments

Wolfram Language, Wolfram Notebook, Wolfram Cloud - welcome to Wolfram Life. To take full advantage of all that Wolfram Community has to offer we require that you sell your possessions and donate the proceeds to the Wolfram Foundation prior to initiation and induction into Wolfram Level 1. It is every Wolfram Student's aim to ascend through the Wolfram Levels and eventually attain the title of Wolfram Master upon which the student will be admitted into the Sacred Order Of Wolfram.

I guess Stephen et. al. noticed that the Jupyter and RStudio type stuff is really a pretty good idea.

edit: ok, i was being too flippant in the above.

I do know the history although as I understand it, the notebook type interface was developed in a few places in the 80s, mathematica being one of them, maple soft another - and all of these influenced by Macsyma and related earlier systems but taking advantage of the improved graphic capabilites as they became available. Not that I was watching this unfold in the early 80s so I might have that wrong.

My flippancy was related to the typical Wolfram self-aggrandizing waffle in the PR approach, and the fact that the most useful part of this (the announced) is arguably deployment to standard web tooling, which a number of at least similar space systems have been doing for the last several years.

Mathematica introduced the notebook UI / language kernel design 30 years ago.

He should open-source this whole thing, create marketplace/appstore for plugins around it for whatever needs to be proprietary.

Not going to happen while Wolfram runs the company.

It's still early days but I'm working on it: https://github.com/wrnrlr/foxtrot

IIRC wasn't one of the design philosophies of Mathematica that everything you need is a part of the core language, eg. you don't need libraries.

He should open-source this whole thing ... and start selling click-bait ads ... with a heavy dose of surveillance ... like the rest of us do.

Notebooks appear first at Wolfram Mathematica and after in Jupyter and RStudio.

I can't tell if you knew iPython / Jupyter were recreations of the 1988 Mathematica Notebook interface, for Python.

https://en.wikipedia.org/wiki/Wolfram_Mathematica

What's the closest equivalent of this, for Jupyter et al?

Is there a seamless, hosted web service for creating & publishing notebooks using open-source tools?

Given how incredibly obvious & inherent to notebooks this use case is, I assume that there must be?

Colab maybe?

In https://observablehq.com/ the platform itself is not open source but the core code that evaluates notebooks is open source, the various libraries used by notebooks are typically open source, and the content of those notebooks is all publicly visible (with license determined by each author).

Aside: Personally I have found that the spreadsheet-like Observable model of tracking a dependency graph between cells and re-rendering dependents whenever a cell changes is much more pleasant than the Mathematica/Jupyter/etc. model where evaluation of cells initially happens from top to bottom at load time and then is afterward haphazard depending on the deliberate action of the viewer (which means it can easily get into inconsistent states).

The resulting notebooks are (at their best) also IMO substantially better as written documents for readers, based on better text layout and richer tools for creating diagrams.

Being JavaScript that runs in the viewer’s browser makes it much easier to build compute-heavy notebooks that rely on reader interaction, and easier for readers to participate as equals with authors.

Just push to GitHub, or any URL for that matter, (Jupyter just have a bit more knowledge about github,gitlab... )

nbviewer.org to view only, mybinder.org to get a temporary docker image in the cloud to run/modify/experiment.

You wan to interact with the gravitational wave LIGO experiement? Here you go: https://beta.mybinder.org/v2/gh/minrk/ligo-binder/master?fil...

You wan t example where the code is hidden voilà has some nice examples.

https://voila-gallery.org/services/gallery/

You need ideas on how to get stated for your language or with different packages here are some minimal demos:

https://mybinder.readthedocs.io/en/latest/sample_repos.html

Nice thing is it work with Python, R, C++, haskell, ... any kernel that Jupyter Run. And you can install any packages in apt/miniconda/CRAN. You want more control ? you can write your own Dockerfile.

You want a single button to publish: Just install a a jupyter extension that for example push to a gist (https://github.com/hadim/jupyterlab_gist)

The big advantage being that you don't _need_ to. If your notebook is already on GitHub well I can make it live if I wish.

There's also this, which I just remembered about:

https://github.com/iodide-project?type=source

Check out CoCalc[1], formerly Sagemath Cloud. It is a project with a lot of history (think of SageMath as python-based open-source Mathematica alternative). It supports Jupyter and Sage notebooks.

CoCalc also supports Julia, LaTeX, R.

I'm using Google Colab[2]. It's the direct equivalent, hosted by Google. Not quite open source, but it's free (and does it matter if a cloud service is open source? If you want control that open source gives you, you run things locally anyway). Disclaimer: I work there, and use it at work daily.

For a Colab vs CoCalc comparison, see[3]. In my brief experience, CoCalc UI is currently snappier than Colab; but I haven't written any large notebooks in it yet.

There are other Python cloud services[4] which I don't have any experience with.

[1]https://cocalc.com/

[2]https://colab.research.google.com/

[3]https://www.forbes.com/sites/gregoryferenstein/2019/08/18/a-...

[4]https://www.dataschool.io/cloud-services-for-jupyter-noteboo...

One thing very much worth plugging about Google Colab is that you get free access to an NVIDIA Tesla K80 GPU.

You're limited to 12 hour stints and you have to keep the window open and network active, but it's enough that I've been able to use it to run neural machine language training (to be fair, I pass model files back and forth, training on EC2 spot instances, too).

CoCalc fully supports publishing Jupyter notebooks to the web [1], which you can see here [2]. The Jupyter notebook renderer that I implemented for CoCalc is a new one written from scratch using React.js, it's is purely client side, and it is faster and more lightweight than the one on GitHub or nbviewer.

[1] https://doc.cocalc.com/share.html#publishing-files

[2] https://share.cocalc.com/share/

take a look at https://nextjournal.com

supports Python, Julia, Clojure, R & import from Jupyter

Indeed, NextJournal's Julia support is quite phenomenal. I like quite a few of the tutorials they have on there.

Is this really new at all? Seems like Jupyter rendering on Github is already a version of this, without expiration of the notebook and having to buy a license to a proprietary product and get sucked into the Wolfram ecosystem. Yes, Wolfram invented the notebook, but just seems like hubris to now claim they have invented putting notebooks on the web. Interesting idea, perhaps a fast follow. Really wish Wolfram would either open source the front end or innovate something that makes it worth spending money on their closed ecosystem.

Two major new releases in a day for notebook computing. A good day!

Reply on news.ycombinator.com