Jupyter Notebook User Experience Survey | Hacker News Reader

Jupyter Notebook User Experience Survey

blog.jupyter.org

70 pointspo8423 comments

What would be really amazing is if Jupyter's notebook server was abstracted further from a general cell-based execution model, and instead provided a generic graphics backend for making web-based GUIs for a variety of programming languages.

Sort of like a combination of GFX.js: https://github.com/clementfarabet/gfx.js/ and PyWidgets, except widgets are no longer constrained to a single cell and the interface is not limited to cells.

This would make it really easy to prototype IDEs and other programming tools in the browser or on native clients - in general it's pretty hard to write an IDE or graphical tool because one needs to integrate system calls to whatever graphics frontend they are writing. However, Jupyter's kernel architecture gives that away for free.

It's a great piece of software.

Starting with Ipython Notebook, but a little unwieldy sometimes. I really hope they keep improving. It seems like all the cool features were introduced early on and then over time there has been almost no progress in the areas I've been using it for.

Some of my criticisms that I hope somone has a good workaround for for now:

Issues I have with it that are not bugs:

- Have to keep a separate firefox profile so that the notebooks don't get lost among other tabs. Maybe a separate built-in Firefox profile or chromium browser would help or maybe its own built in browser that would make it more of an IDE.

- Stores data frames and such in memory. Not clear which variables are taking up all the space. I guess restart the kernel... This is not as much of a problem with Ipython because those sessions tend to be more ephemeral. So tracking of variables by memory usage would be wonderful.

- Freedom to edit any cell results in a bit of a mess quickly with out of order execution of things. Probably some kind of organizational strategy they could encourage through the UI would be great. Otherwise, reading others' notebooks is impossible unless they polished it for some time.

- Scrolling is really bad if a command created a lot of output.

- Searching through worksheet is not intuitive. Always have to look up ipython manual.

Issue that I think is a bug and is the major reason I'm not using it more often:

- Cannot reload modules easily. %load_ext autoreload %autoreload 2 doesn't work. %autoreload 1 doesn't work.

Tested on OSX and Linux.

Sometimes after messing with it for a while it starts working for a bit but then segfaults. I should look into this more.

Same problem with ipython, but this is more of an issue with the notebook because it has more of a permanent state than terminal sessions. You end up having to restart the kernel and lose everything in memory, which is the main reason to use the notebook in the first place.

It would be really cool if there was a feature that let you know when you modified a function but are using the result of it that was obtained before modifying it. Or if you modified a module but are using the old version.

re reloading modules: I eventually found it easier to simply %run external code, rather than trying to import and reload modules.

> separate firefox profile...

Is it possible to run it in a tab-less chrome/ff instance through arguments? If not, this should be possible.

Jupyter Notebooks are great, I use them a lot for scientific computing and data analysis.

The critical missing features in my opinion are 'undo' and proper integration with Git (or another VCS).

So what is the usual usecase for these Jupyter notebooks? I see everyone praising them, but I still don't really know a good use case. I thought about using it as notebook for work stuff, but I don't want to run something in command line and then in browser to make a simple note when I can just open vim and make a quick note

For me: anything involving data (data scraping, exploratory data analysis, plotting, testing various machine learning techniques, etc); here short feedback loop is crucial + it's great to have plots in place.

Also (a bit) for playing around with new libraries / databases etc. (Much better than both Python scripts (need to be rerun from scratch) and Python shell.)

For me, it's nice for interactive code and visualization development. Particularly when parts of a script are slow running, being able to do it once and then iterate on a dependent part in a notebook cell saves a lot of time.

Once I'm done, I can document it inline with markdown and latex.

Recent use for me:

Semi-technical boss comes to me and says "Hey look! I wrote this cool SQL query to figure out what time of day our service is being used most. Can you turn it into a heatmap?"

And of course you know if you do something like this once, you will be asked to do it again, until pretty soon you will be expected to do it automatically every week or something.

So: Jupyter notebook to the rescue. Query the DB, wave hands a few times, produce a heatmap (total lines of python: about 10). Then give boss the notebook and tell him if he wants to update or modify, have at it, here's where your query is.

I never have to deal with it again, it's self-documenting and transparently future-friendly (anybody with an ounce of sense can edit it) and it looks all mathematica and stuff. Case closed.

Exploratory data analysis. I may need to play around with loading data from a format I'm unfamiliar with, then do some basic summaries or information gathering about the data. This makes it straightforward to iterate on this type of stuff (instead of having to use command line REPL and press up arrows alot to re-run code sections without re-typing or pasting code). It's straightforward to share with someone familiar with Jupyter, and to a lesser extent with others who just need to look through the analysis results or code shown by the notebook. Though to be honest, when there isn't something about the analysis Python makes simpler, I find R with RStudio to have more flexibility for this kind of stuff, personally.

Verification. If you compare traditional data analysis in Excel and such, it is much easier to verify what is going on in a notebook than it is in multitudes of Excel sheets. Easy calculation views, graphs, tables, etc.

Jupyter is awesome. However, something about the orange in the colour theme just makes the interface seem so dull to me.

try Beaker: http://BeakerNotebook.com we have a modern interface & the next release will have a dark color theme too, easily accessible from a menu item: https://twitter.com/beakernotebook/status/682233629460795393

I've heard it a few times on HN (are you its dev?). Can you name so advantages of Beaker over Jupyter?

1) modern interface 2) polyglot architecture (multiple languages work together in the same notebook) 3) native electron version

yes i am one of its developers.

here's an example of working with Python and D3/JavaScript together: https://pub.beakernotebook.com/#/publications/560c9f9b-14e6-...

and the electron version: https://github.com/twosigma/beaker-notebook/wiki/Electron-Be...

You can install custom CSS somewhere in your .jupyter directory. Of course, this feature is undocumented and different from the analogous (documented) ipython feature.

Check this, it might be helpful for you https://github.com/merqurio/jupyter_themes

Sorry, silly question:

I've been out of the loop for some time it seems -- what is this 'Jupyter'? It looks like it's basically ipython... how is it different from ipython? Is it ipython2.0 in spirit or something?

They took the notebook part from IPython and made it language agnostic, this is now Jupyter. You can use Jupyter Notebook together with many languages (Jupyter Kernels) such as R, Julia, Bash and many more in addition to Python.

Ahhh... thank you for explaining.

Including Bash, huh? I remember I used to think that if only a few things were different in ipython I would actually use it as my main terminal window. With Jupyter running bash, are people using it as their main terminal now?

Running a Bash kernel in Jupyter Notebook might be convenient if you are running your Jupyter server on a remote machine. So you don't have to SSH into the server for some quick changes, you can stay in your browser. I haven't used it myself as I only use it locally.

So actually the main reason I'm interested in ipython-style notebooks for my terminal shell needs is the benefit of having... graphs! and pictures! and other graphical things in a terminal shell setting. I'm willing to put up with inconveniences you speak of if I can get the graphical things working well. People have experimented with it, it seems: http://jeroenjanssens.com/2015/02/19/ibash-notebook.html I'll look more into it now.

Only for expository processing, keeping things transparent for people who are not necessarily programmers. Still waiting for spell checking in markdown blocks

Reply on news.ycombinator.com