Three months after part 1, the second half of the UCL Software Carpentry workshop for High-Energy Physics took place on Friday 15th February 2013.
The experimental format with two sessions three months apart might have worked better if, as envisaged, we had managed to get an online discussion going in the gap. Unfortunately this did not work out. Despite several attempts at chasing up the participants, no-one registered or posted anything on the course web site. After talking to some of the students, I think their interest was just not great enough to overcome the bureaucracy involved in registering on the UCL Moodle server and the tendency to focus on their other (assessed) courses and research projects. Attendance was down from 65% to 30% (6 of 20 students), although those who did attend were clearly motivated, worked hard and asked thoughtful questions. With two helpers (James Hetherington and HEP postgrad Sam Cook) and myself as instructor, we had a student:staff ratio of 2:1.
Most of the morning session was spent getting back to where we were at the end of part 1, with various technical problems needing to be solved all over again. Surprisingly, everyone came back after lunch and the afternoon session was, I think, rather more successful. Taking advantage of the small student numbers, we divided the class into two teams of three, and set them to work on different parts of the same code base. One team wrote code to retrieve a list of particle four-momenta from a collision event, while the other team worked on generating a list of oppositely-charged pairs from a list of particles.
Due in part to the nature of their tasks and the input they had to work with, the first team took a fairly ad-hoc approach to testing their code, while the second wrote a fairly comprehensive set of unit tests. In between iterations (of about 20 minutes each) they talked about what they had done and what problems they had encountered, and after each iteration they pushed their code to the shared repository on GitHub, merging their changes where needed.
During the last half hour I pulled the merged code from both teams into my local repository and used a separate Git branch (based on my own version of the code) to plot the results. Then I rushed through a list of topics we had not managed to cover, including testing floating-point numbers.
While there is a lot more material I would have liked to cover, and could perhaps have covered by imposing a more rigid structure on the event, I suspect that the most effective learning experience was the rather chaotic extended exercise in collaborative code development. What I would like to do next year is find a way to scale this up to what I hope will be a larger class.