In-depth comparison of cloud-based services that run Jupyter Notebook

blog-post

#1

Hello all. I’m nearly finished writing an in-depth comparison (~5000 words) of six different services that allow you to run your Jupyter notebook in the cloud for free. One of those services is Binder, of course!

I’ve spent at least 40 hours on the research and writing process, and I believe that I’m accurately portraying each of the six services. However, before I publish I’m reaching out to each of the services to see if they will help me to verify the accuracy of my article.

I have two specific questions (plus a request):

Question 1:

One of the criteria I’m using to compare the services is which languages it supports. Is it accurate to say that “any programming language supported by Jupyter can be used with Binder”?

  • If yes: Is this kernels list the most current list of languages supported by Jupyter?
  • If no: Is there a complete list of languages that can be used with Binder?

Question 2:

The Binder FAQ states:

Binder will automatically shut down user sessions that have more than 10 minutes of inactivity (if you leave your window open, this will be counted as “activity”).

Is that an accurate statement? My finding is that leaving the Binder browser tab open (without a running computation) does not seem to count as “activity”, and sessions do shut down automatically (after perhaps 30 minutes, though I haven’t timed it).

Additional request:

Would anyone be interested in reading my draft of the Binder section to provide additional fact checking? Basically, I would send you (via private message) the introduction to the blog post plus the section about Binder, and you can read through it to see if there are any statements that are incorrect or misleading. This would require reading ~1000 words which would probably take 5 to 10 minutes.

Regardless of whether anybody is able to help, I will surely reply to this topic with a link to the article once it is live on my blog!

Thanks very much! :raised_hands:


#2

Per @minrk December 21st , 2018 8:03:

“The docs shouldn’t say that leaving a window open counts as activity (this used to be true, but is not anymore).”


#3

Yes, if you can install it in Docker. Worse case you will have to write your own Dockerfile.

Is this kernels list the most current list of languages supported by Jupyter?

This is community maintained, we trust the community, so it is “supported” in the sense that if the user put it there, they think it should work.

  • If no: Is there a complete list of languages that can be used with Binder?

I don’t believe.Note that binder does not only support Jupyter and Jupyter Kernels, there are prototypes (demos?) that use open-refine, Rstudio…

of six different services that allow you to run your Jupyter notebook in the cloud for free.

Would you mind sharing this list ? Just to potentially give you a list of others ? I can think of:


#4

@fomightez and @carreau - Thanks so much for answering my questions!

The services I included in my comparison are Binder, Kaggle Kernels, Google Colaboratory, Azure Notebooks, CoCalc, and Datalore. (Datalore uses its own file format and is the furthest from the classic Jupyter Notebook interface, but it met all my criteria for inclusion, including that you can import/export the ipynb format.)

I didn’t include any services that only give access to the JupyterLab interface (such as Notebooks AI, Kyso, and CyVerse).

I didn’t include IBM Watson Studio Cloud because the process of setting up was cumbersome, the interface is complicated, and I ran into lots of errors during my testing.

I didn’t include any paid-only services (such as Crestle.ai, Paperspace, and Salamander).

Thanks so much for your list! Since I’m only comparing free services, that excludes SageMaker, R-Brain, and Datalab (which is free but incurs “usage” costs). Gryd has a free plan, but only for people with an academic email address. Code Ocean is similar, in that the free plan is severely limited unless you have an academic email address. All of these are potentially excellent options for some people, but my particular comparison is focused on free services that anyone can use.

Thanks again! If anyone decides they would like to give my draft of the Binder section a read, let me know. Regardless, I’ll post here once the article is live!


#5

happy to give the binder section a read!

one thing I’ll note: while all those in your list are free, Binder is the only one (maybe with the exception of CoCalc?) that is an open project, and that isn’t attached to the business model of a single company. Something I think is important!


#6

Thanks @choldgraf! That’s a great point, and I will mention it in the article in the “how to choose” section.

I will send you a draft via private message shortly.


#7

I’ve got a mess of a list some notebook hosts in a mess of a wiki here:

Several online workbench environments provide access to notebook interfaces:

Data Science workbenches w/ notebook integration

Workbench repos / Jupyterhub Integration

Desktop / Electron App notebook environments

  • Stencila (also available as an HTML web service, eg via MyBinder)
  • nteract

On my to do list is figure out a proper classification / taxonomy for stuff I want to try to log / track and create that structure in the repo proper (rather than the wiki).

–tony


#8

Hello this thread!

One more for the list:
https://www.illumidesk.com/


#9

Thanks for writing a blog like this! Just reading the questions you asked makes me excited about the whole post :slight_smile:

I think it is tricky to sort this by “language” (at least for Binder). You can run software written in many languages (there is a terminal where you can run any command you want after all). This means that it is maybe better to ask “what language/software can be used given that the UI has to live in a browser”. For example you can have VS Code on Binder https://twitter.com/betatim/status/1106195567892402177 and run things through that.

I’d be happy to read your draft @dataschool.


#10

Awesome! I am excited to share it with the world!

Thanks! I will send you a draft via private message shortly.


#11

I just discovered https://www.saturncloud.io/ which also seem to offer hosting of Jupyter notebooks.


#12

ModeAnalytics is a data science platform running Jupyter Notebooks and specializing in relational database integration. It’s free to use with a paid version for teams and extra reporting options. They’ve been around for a while.


#13

I would very much like to read it as well.


#14

@fm75 Thanks for your interest! At this point I don’t need any more folks to read the draft, but I’ll be publishing the finalized article next week, and will definitely share a link to the article! :link: