Last week we ran an R-based SWC bootcamp at University of Sydney. The primary target audience encompassed grad students from the Department of Psychology who had little to no experience in programming. While some had previous limited experience with programming languages (e.g. MATLAB or R), most of them had been exposed only to statistical software such as SPSS.
The audience was beginners, mostly from the Department of Psychology. The instructors were Dan Warren (a post-doc in Biological Sciences at Macquarie University) and Diego Barneche (a PhD student at the same university), and our host was Prof. Alex Holcombe (Dept. of Psychology, University of Sydney). The material we covered included:
Day 1 Basics of R programming, including data types, subsetting and vectorized operations; functions in R; How to structure and organize projects; control flow (if/else, for and while loops, apply family in R, and the package plyr).
Day 2 The Unix shell; version control in Git; exercises on GitHub; reproducibility, including an example with the package knitr.
Most of the content was well managed within the original schedule. We used RStudio because it is a great user-friendly interface for beginners. Also, we provided them with a .zip file at the beginning of the bootcamp. This file contained a folder for each lesson, where an .RProj file with initial scripts and data was already provided. This approach confers nice advantages as each lesson already serves as an example of self-contained project, every student is standardized in terms of initial material and we avoid problem with full vs. relative paths.
On the first day (Friday), we had 28 students. The room had two screen projectors available. One had RStudio opened 100% of the time, and the other contained Etherpad or some webpage used as an example to illustrate a case. On the second day (Saturday), we had only 15 students, but we had to move to a different room with one projector, so it was a bit annoying to switch between applications and/or tabs (e.g. from RStudio and Etherpad) and adjust screen resolutions when Dan and I had to swap our laptops.
Diego: I suspect that the reason why we had a considerable drop-out from day 1 to 2 involved a combination of three things: (1) because it was Saturday; (2) because many students couldn't keep up with the pace on the first day; and (3) because they didn't have to pay for the course, so it wouldn't be a big deal to drop at a late notice. In my perception cause (2) was the major one, and I finally realized how hard it is to teach novices how to program. Maybe I spent too much time on the basics, including a bunch of examples on data structures in R. Although some people would rather drop some of the basics, I feel like without a basic foundation of data structures, R learners won't have the tools to teach themselves in the future and speed up their learning experience.
I wonder if a three-day bootcamp is something to be reconsidered when it comes to novices. I do understand that people's (both instructors and students) availability is a major issue here, but I feel sometimes frustrated that we need to cover too much in a short amount of time. For instance, beginners have many questions regarding functions, and the topic itself is worth an entire afternoon (at least) of coverage. In such an ideal scenario, I would have chosen to cover basics and project set up during the morning of the first day, and then spent the entire afternoon teaching functions. Second day would be dedicated to control flow in the first half and shell in the second. And the final day dedicated to git, exercises on GitHub (including branching, push/pull, merging conflicts and pull requests) and reproducibility.
In any case, the audience was great in general, and many students took great advantage of the course. The git lessons were really nice, and students seemed to have fun with push/pull exercises and merging conflicts using GitHub. After having participated in 4 bootcamps this year where shell was not the primary language, I wonder how much we need it. I'm not entirely sure that students are immediately convinced it is a great tool to operate a computer (though I use shell on a regular basis for my own work). Maybe part of it relates to all the unpredicted bugs on Windows machines - this can always be super annoying, or even its relatively reduced functionality on Windows. Maybe part of it relates to the fact that we don't have much time to cover it, so things like for loops, functions (covered in R anyway) and pipelining, which are the coolest topics in shell in my opinion, have to be put aside.
Dan: When teaching functions, I really tried to use a lot of simple in-class exercises, and I feel like that was helpful. I think in the future I'm going to work towards including even more exercises, as I feel like it gives both the students and the teacher a moment to calibrate their retention and understanding of the material.
I feel like everyone followed along with the shell exercises, but after the session I was asked by more than one student why they should bother to learn this. I feel like we came up with a couple of decent answers for them, but it really made me think; R is by its nature going to be of interest to a lot of scientists who are specifically interested in analysis, but who are not interested in becoming super-elite programmers. I've never sat in on one of the Software Carpentry Python courses, but I wonder if it's generally easier for them to motivate students to get into the shell and version control. Whether that's true or not, it really did emphasise for me that motivation is one of the key starting points for learning, and that imparting motivation is essential to good teaching. Something to think about for next time!
Please leave your comments below, we'd really appreciate some general feedback from everyone who has (and hasn't) been on a similar situation.