Teaching basic lab skills
for research computing

Reflections on Claremont

The Claremont Colleges bootcamp just wrapped a couple of hours ago, and judging from my pile of happy sticky note responses, I think we nailed it pretty good. Things were a bit fast on the first day of intro shell and Python, but I think we hit something pitch perfect for day two on intermediate Python and Git. As this Python-heavy workshop progressed, some questions started coming into focus from student responses, discussions with instructors and organizers, and examinations of just how to teach this ascendent skill in scientific computing. To be clear: I have never taught our Python material, and I consider myself a fairly casual Pythonista. This is a debate that the masters need to consider, but I'm hoping here to use my agnostic outsider's perspective to help set the conversation up.

Two major axes span the space of teaching Python, and actually they're just the same two big questions found in teaching anything: what to teach, and how to teach it. But these old pedagogical chestnuts really resonate with the practice of teaching Python, because the place that Python's ecology has evolved to has made them pressing and polarized matters.

First up: what to teach. I've had the privilege of watching two awesome deliveries of the Python material at the two bootcamps I have taught, but excellent as they were, they could not have been more different. The Claremont Python lecture began from a place of absolute fundamentals—what is a variable type, what is assignment, what is an operator—those omni-linguistic channels of thought that shape programmatic logic and structure everywhere. Approaching the problem not just as basic Python but basic programming laid down a bulletproof foundation that moved gradually enough and depended on little enough chuft that almost everyone could keep up, even with a huge and varied group like we had at Claremont.

Python in Victoria earlier this year was a different animal altogether. That workshop was heavily tools oriented—lots and varied plotting, calculating some simple statistics from datasets using things like numpy and matplotlib (as I recall), and heavy use of the IPython Notebook. Actually, I'll never forget one quasi-magical thing that happened during the Victoria Python lecture; it was time to introduce for loops, and I expected the instructor would get bogged down picking apart syntax and trying to hash out in exact detail the guts of this pattern that students wrestle with the first time they see it; but instead, they just sailed right through it, talking about it like it was second nature—and the students got it! Somehow they assimilated this new idea with barely a speedbump, and just rolled right along playing with plotting tools. Waltzing through the chuft and getting students to get their hands dirty with useful tools right away seems so compelling and engaging, especially when they walk out of the room agog with possibility, but one person made a point I can't shake: how demoralizing is it for students when the next day they can't use any of the tools they were so excited by, because they don't have enough fundamentals to proficiently wield them?

Anyway, the proof is in the sticky notes, and much to my surprise, there wasn't really much of a criticism of pace or level of the Python material at Claremont on day one—the comment that got repeated a bunch of times, was "I wish there had been more hands on examples and puzzle problems to work on". So where is that shining kernel of value here—in the self-actualization and engagement of hands-on activity, or in the exciting promise of all the cool things the world of Python has to offer? Is it even possible to make engaging examples out of the tiniest nuts and bolts of programming, and can we float the nifty packages while imparting enough fluency that students can drive them on their own? The "what to teach" axis of this problem space can be reskinned as a slider from 'foundations' to 'tools'; finding the happiest home therein is to my mind an open question.

The second axis was of course, "...and how to teach it". The choice of vehicle for how to introduce Python seems to me like a pretty crowded room. You've got your standard vanilla.py to run at the bash prompt; you've got your command-line IPython shell; you've got your IPython Notebook; and you've got your web-based Python environments—and each one has its own virtues and pathologies. Where we really ran afoul of this embarrassment of riches at Claremont, was when we tried to use scripting and the interactive shell all at the same time. For many of these students, this was their first exercise in anything REPL, and their first dive into raw text document creation; all these boxes with nothing but text looked kind of the same to them after a while, and it showed in the questions and comments. Many of the helper questions during the day boiled down to students getting confused between text editor, IPython prompt, and bash prompt, and that same confusion was explicitly called out in the sticky note feedback at the end. It was glaring clear that we needed to clear the air, and give the students something solid to focus on, instead of asking them to rapidly switch their thinking from scripting to bashing to interactively issuing Python commands—all things they had encountered for the first time in the last couple of hours.

After this experience, the IPython Notebook's star seemed to be rising. Here is a one hundred percent contained environment where code is entered inline and results are delivered immediately afterwards both spatially and temporally (an instant and context-heavy feedback that is so key for learning). But problems lurk here too. As I was helping for the second Python lesson at Claremont (which used the notebook exclusively), at least half the strategies I gave boiled down to "hit run all and it'll probably be okay". Students were getting lost in the nonlinear state machine that the notebook implicitly creates, without getting a chance to get a solid grasp of what that meant or even that that was the case. We were able to keep on rolling just by re-running every cell, but there wasn't time for the students to really reflect on what that was all about. All this was, of course, after scrambling to get everyone's notebooks working in the first place; in terms of overhead, the notebook is an aircraft carrier compared to the log raft of a script to run at the command line—immensely powerful but sometimes tricky to bring together.

It wouldn't be a post from me if I didn't consider an idea probably no one will like: what if we cut way down on the big guns, and stuck to a script-and-execute model? This would not only eliminate the tragedies of getting all the IPython and Notebook stuff working for everyone, but would start building the muscle memory of writing programs and running them at the command line, which is much more reflective of the day-to-day trade of building software. Getting some of these extra platforms out of the way might make room for fundamentals as well as exciting and powerful package and tool demonstrations, and keep the whole process much more focused, which is one thing we absolutely need to do.

And yet, there is absolutely no pedagogical equal for immediate feedback, which is something the script-and-run model is slow to yield, while IPython nails it cold. So as a middle ground, what about an IPython Notebook where we encourage students to run the whole book each time they added to it, and maybe take some time to point out how stuff gets weird if you execute cells out of order. This isn't too far a cry from building a complete and consistent script, it keeps the student's attention all in one place, and it provides compelling immediate feedback; set it up so there's a 20 minute example at the start that is non-fatal to miss for those still struggling to get the notebook up, and we might have a way through the crowded room. That's one untested strategy—but all told, we've got another slider to adjust in our approach to teaching Python: rock-solid programs, or dynamic interactive environments?

Where you make contact with the dartboard of content vs. style will largely boil down to your priorities and personal flare; since the full range of these options is valuable, our goal should not be to pick winners and losers, but to pull in the best of all of what's out there. What the data seems to be showing after this latest round, is keep your platform focused, and keep it very hands-on.

Dialogue & Discussion

You can review our commenting policy here.