What's the bad news?

There are still a few places where the ipython notebook is either not ready or simply not a good solution.

UNIX state and ipynb state can differ: Over the summer I found that people new to UNIX and programming got very easily confused by the fact that ipynbs were distinct from the underlying state of the machine. Experienced computer folk would automatically understand that if three notebooks were running on a single machine, then file changes in one would be visible to other notebooks. The notebook interface, however, masks this underlying change. I think this is unavoidable.

Underlying ipynb state isn't linear in the notebook: A related problem is that ipynb state doesn't progress "down" the notebook, because you can always jump around in the cells. I think this is unavoidable, too.

Installation problems: Unless you buy into the Enthought Python Distribution, IPython Notebook and its "useful" dependencies, which include matplotlib, are basically impossible to install. Even when you do use the EPD, about a quarter of the installations don't work properly right off the bat, and I have generally not been able to debug things in the time available to me. This is the main reason I'm using EC2, and why I'm going to encourage people in my class to switch to VMs in the future; installation is easier that way.

Long-running jobs: IPython Notebook doesn't effectively run in a headless mode, and excepts you to be more or less tethered to the same Internet connection. This makes it hard to execute long-running jobs in ipynb, because you can't close up your laptop and come back later very easily. There are exceptions to this but not very comfortable ones. As a general rule, I do all of my long-running compute at the command line. (Note, Fernando says they have plans to improve upon this aspect of the notebook.)

Writing serious code: I don't think the ipynb is going to replace my serious editor (emacs) for writing modules and bigger programs (more about this later). It's meant for interactive data exploration.

It gets in the way if you already kinda know UNIX: One of the most common complaints I got at the summer course was that people who already knew some UNIX didn't like IPython Notebook for anything but plotting. They already knew how to drive the operating system and didn't need a journal-like thingy to help them.

I'm pretty ambivalent about counting this last one as a negative -- I suspect that at these students become more familiar with UNIX they will be able to more easily navigate between ipynb and UNIX, but at an early-intermediate stage of knowledge it gets in the way. More generally, I think that we should be teaching scientists to use this kind of journal-like thinking when they do computation. But that doesn't necessarily make it easer on them when they're learning it!

Now, what's the good news?

Ease of use: I think the best news is that the IPython Notebook is easy to use and easy to understand at a basic level. No more remote command lines, remote editors, and file transfer -- you can learn Python to a basic level without all that stuff. My experience has been that, in biology, throwing all of that at people with no prior computing experience is just too much. I've been positively impressed with how quickly my students have been picking up the programming side of things, and I honestly expect the command line stuff to be much more straightforward. (We'll see how the next few weeks go!)

Very easy for demos: I've been using it in screencasts and in-class demos for my graduate course, and some colleagues here at MSU have been using it for in-class demos for their intro programming course. Thumbs up. We can pass the code around, or post static views of notebooks, or whatever. Overall, the ability to give the students exactly the code I worked up in class is golden.

Good data exploration: I'm increasingly starting to use it to whip together graphs and figures for talks; we've already used it for publications, of course (here), but it's a great way to interactively explore data sets.

nbviewer is awesome: One of my gripes in the original teaching post was that it was hard to post static notebooks. No longer! The nbviewer.ipython.org site lets you transform raw .ipynb files (from github gists or any URL at all) into nicely rendered HTML notebooks.

IPython Notebook is not a sandbox: In my experience, you can go quite far in data exploration with cell-level functions and scripting. Until you become reasonably expert at those, you probably don't need to write much in the way of your own modules. Moreover, even once you do start writing modules, ipynb lets you work with those via the standard Python systems.

More broadly, IPython Notebook is a fantastic user interface on top of the increasingly broad and deep Python ecosystem for scientific computing and data analysis. It is not, in any way, a sandbox that limits you or prevents you from making full use of what other people are doing. And that, from the perspective of teaching the actual practice of scientific computing and data analysis to students, is what I find most important.

How am I actually using it?

In my intro computational science course: In the first few weeks, students learn how to start up an EC2 instance (YouTube), connect to IPython Notebook (YouTube), do basic Python commands, upload their homework to github gists as IPython Notebooks (YouTube), and do many useful things like plot (YouTube). I post the homework sets as ipynbs, too -- see, for example, the Monty Hall problem.

For the summer course on next-gen sequence analysis: I post a bunch of notebooks through github that show students how to update their notebook repositories, install software, run BLAST, plot data, plot k-mer distributions, access sequence collections programmatically, and even play with digital normalization. Next year I plan to put in more info on what's going on in the notebooks -- this year, I talked through them interactively in front of the class.

It's important to note that teaching all this stuff is still the primary challenge, and the IPython Notebook is merely one tool that we can use. Still, it's a pretty awesome tool, given the craptitude we have in terms of legacy development environments, and the reviews have been pretty positive so far from the students.

This next year at PyCon we've proposed a panel to talk about the ipynb for teaching, and I think I'll be pimping it at a pre-PyCon teaching workshop, too.

What does the future hold?

When I talk to the IPython team, I find them to be incredibly ambitious. They basically view the IPython Notebook as a general computing platform for seamlessly connecting a Python interpreter to dynamic HTML/JavaScript, and they are hell-bent on the awesome. I confidently expect to see generic JavaScript widgets, spread-sheet like computing, collaboration within an ipynb, slide shows, and recording/playback to make an appearance within the notebook over the next year or two.