Posted: June 24, 2014
The first talk, “Things Coming Together: Learning Experiences in a Software Studio”, was presented by Julia Prior, from UTS. (One of the nice things about conferences is catching up with people so Julia, Katrina and I got to have a great chat over breakfast before taxiing into the venue.)
Julia started with the conclusions. From their work, the group have evidence of genuine preparation for software practice, this approach works for complex technical problems and tools, it encourages effective group work, builds self-confidence, it also builds the the more elusive prof competencies, provides immersion in rich environments, and furnishes different paths to group development and success. Now for the details!
Ther are three different parts of a studio, based on the arts and architecture model:
- People: learning community
teachers and learners
- Process: creative , reflective
- physical space
- Product: designed object – a single focus for the process
UTS have been working on designing and setting up a Software Development Studio for some time and have had a chance to refine their approach. The subject was project-based on a team project for parks and wildlife, using the Scrum development method. The room the students were working in was trapezoidal, with banks of computers up and down.
What happened? What made this experience different was that an ethnographer sat in and observed the class, as well as participating, for the whole class and there was also an industry mentor who spent 2-3 hours a week with the students. There were also academic mentors. The first week started with Lego where students had to build a mini-town based on a set of requirements, with colour and time constraints. Watching the two groups working at this revealed two different approaches: one planned up front, with components assigned to individuals, and finished well on time. The other group was in complete disarray, took pieces out as they needed it, didn’t plan or allocate roles. This left all the building to two members, with two members passing blocks, and the rest standing around. (This was not planned – it just happened.)
The group that did the Lego game well quickly took on Scrum and got going immediately, (three members already knew about Scrum), including picking their team. The second group felt second-rate and this was reflected in their sprint – no one had done the required reading or had direction, thus they needed a lot of mentor intervention. After some time, during presentations, the second group presented first and, while it was unadventurous, they had developed a good plan. The other group, with strong leadership, were not prepared for their presentation and it was muddled and incomplete. Some weeks after that presentation practice, the groups had started working together with leaders communicating, which was at odds with the first part of the activity.
Finding 1: Group Relations.
- Intra-Group Relations: Group 1 has lots of strong characters and appeared to be competent and performing well, with students in group learning about Scrum from each other. Group 2 was more introverted, with no dominant or strong characters, but learned as a group together. Both groups ended up being successful despite the different paths. Collaborative learning inside the group occurred well, although differently.
- Inter-Group Relations: There was good collaborative learning across and between groups after the middle of the semester, where initially the groups were isolated (and one group was strongly focused on winning a prize for best project). Groups learned good practices from observing each other.
Finding 2: Things Coming Together
The network linking the students together doesn’t start off being there but is built up over time – it is strongly relational. The methodologies, mentors and students are tangible components but all of the relationships are intangible. Co-creation becomes a really important part of the process.
Across the whole process, integration become a large focus, getting things working in a complex context.Group relations took more effort and the group had to be strategic in investing their efforts. Doing time was an important part of the process – more time spent together helped things to work better. This time was an investment in developing a catalyst for deep learning that improved the design and development of the product. (Student feedback suggested that students should be timetabled into the studio more.) This time was also spent developing the professional competencies and professional graduates that are often not developed in this kind of environment.
(Apologies, Julia, for a slightly sketchy write-up. I had Internet problems at the start of the process so please drop me a line if you’d like to correct or expand upon anything.)
The next talk was on “Understanding Students’ Preferences of Software Engineering Projects” presented by Robert McCartney. The talk as about a maintenance-centrerd Sofwtare Engineering course (this is a close analogue to industry where you rarely build new but you often patch old.)
We often teach SE with project work where the current project approach usually has a generative aspect based on planning, designing and building. In professional practice, most of SE effort involves maintenance and evolution. The authors developed a maintenance-focused SE course to change the focus to maintenance and evolution. Student start with some existing system and the project involves comprehending and documenting the existing code, proposing functional enhancements, implement, test and document changes.
This is a second-year course, with small teams (often pairs), but each team has to pick a project, comprehend it, propose enhancements, describe and document, implement enhancements, and present their results. (Note: this would often be more of a third-year course in its generative mode.) Since the students are early on, they are pretty fresh in their knowledge. They’ll have some Object Oriented programming and Data Structures, experience with UML class diagrams and experience using Eclipse. (Interesting – we generally avoid IDEs but it may be time to revisit this.)
The key to this approach is to have enough projects of sufficient scope to work on and the authors went out to the open source project community to grab existing open source code and work on it, but without the intention to release it back into the wild. This lifts the chances of having good, authentic code, but it’s important to make sure that the project code works. There are many pieces of O/S code out there, with a wide range of diversity, but teachers have to be involved in the clearing process for these things as there many crap ones out there as well. (My wording. 🙂 )
The paper mith et al “Selecting Open Souce Software Projects to Teach Software Engineering” was presented at SIGCSE 2014 and described the project search process. Starting from the 1000 open source projects that were downloaded, 200 were the appropriate size, 20 were suitable (could build, had sensible structure and documentation). This takes a lot of time to get this number of projects and is labour intensive.
Results in the first year: find suitable projects was hard, having each team work on a different project is too difficult for staff (the lab instructor has to know about 20 separate projects), and small projects are often not as good as the larger projects. Up to 10,000 lines of code were considered small projects but theses often turned out to be single-developer projects, which meant that there was no group communication structure and a lot of things didn’t get written down so the software wouldn’t build as the single developer hadn’t needed to let anyone know the tricks and tips.
In the second year, the number of projects was cut down to make it easier on the lab instructors (down to 10) and the size of the projects went up (40-100k lines) in order to find the group development projects. The number of teams grew and then the teams could pick whichever project they wanted, rather than assigning one team per project on a first-come first-served approach. (The first-come first-served approach meant students were picking based on the name and description of the project, which is very shallow.) To increase group knowledge, the group got a project description , with links to the source code and commendation, build instructions (which had been tested), the list of proposed enhavements and a screen shot of the working program. This gave the group a lot more information to make a deeper decision as to which project they wanted to undertake and students could get a much better feeling for what they took on.
What the students provided, after reviewing the projects, was their top 3 projects and list of proposed enhancements, with an explanation of their choices and a description of the relationship between the project and their proposed enhancement. (Students would receive their top choice but they didn’t know this.)
Analysing the data with a thematic analysis, abstracting the codes into categories and then using Axial coding to determine the relations between categories to combine the AC results into a single thematic diagram. The attract categories were: Subject Appeal (consider domain of interest, is it cool or flashy), Value Added (value of enhancement, benefit to self or users), Difficulty (How easy/hard it is), and Planning (considering the match between team skills and the skills that the project required, the effects of the project architecture). In the axial coding, centring on value-adding, the authors came up with a resulting thematic map.
Planning was seen as a sub-theme of difficulty, but both subject appeal and difficulty (although considered separately) were children of value-adding. (You can see elements of this in my notes above.) In the relationship among the themes, there was a lot of linkage that led to concepts such as weighing value add against difficulty meant that enhancements still had to be achievable.
Looking at the most frequent choices, for 26 groups, 9 chose an unexacting daily calendar scheduler (Rapla), 7 chose an infrastructure for games (Triple A) and a few chose a 3D home layout program (Sweet Home). Value-add and subject-appeal were dominant features for all of these. The only to-four project that didn’t mention difficulty was a game framework. What this means is that if we propose projects that provide these categories, then we would expect them to be chosen preferentially.
The bottom line is that the choices would have been the same if the selection pool had been 5 rather than 10 projects and there’s no evidence that there was that much collaboration and discussion between those groups doing the same projects. (The dreaded plagiarism problem raises its head.) The number of possible enhancements for such large projects were sufficiently different that the chance of accidentally doing the same thing was quite small.
Caveats: these results are based on the students’ top choices only and these projects dominate the data. (Top 4 projects discussed in 47 answers, remaining 4 discussed in 15.) Importantly, there is no data about why students didn’t choose a given project – so there may have been other factors in play.
In conclusion, the students did make the effort to look past the superficial descriptions in choosing projects. Value adding is a really important criterion, often in conjunction with subject appeal and perceived difficulty. Having multiple teams enhancing the same project (independently) does not necessarily lead to collaboration.
But, wait, there’s more! Larger projects meant that teams face more uniform comprehension tasks and generally picked different enhancements from each other. Fewer projects means less stress on the lab instructor. UML diagrams are not very helpful when trying to get the big-picture view. The UML view often doesn’t help with the overall structure.
In the future, they’re planning to offer 10 projects to 30 teams, look at software metrics of the different projects, characterise the reasons that students avoid certain projects, and provide different tools to support the approach. Really interesting work and some very useful results that I suspect my demonstrators will be very happy to hear. 🙂
The questions were equally interesting, talking about the suitability of UML for large program representation (when it looks like spaghetti) and whether the position of projects in a list may have influenced the selection (did students download the software for the top 5 and then stop?). We don’t have answers to either of these but, if you’re thinking about offering a project selection for your students, maybe randomising the order of presentation might allow you to measure this!