A couple of years ago, while I was teaching, I wrote some software to try and schedule lab sessions for 100 second-year students at Swansea University, with various constraints. Previously this was done by shuffling around boxes in Excel until they fit. I was sure that a computer could do better in less time.
If you ignore the many hours I spend writing the software, I was right about the time it would take. Sadly I never managed to get better than "about as good" as could be managed by hand. However, other huge improvements were achieved: including personalised schedules, rather than having to read through an intricate table, and assessment barcodes so that three separate cover sheets were no longer needed.
This was my first "big" project in Python, so please forgive the poor structure. I'm releasing it:
- in case any part of it is useful to anyone, although the scheduling algorithm isn't great and a lot of the things it does are very bespoke to Swansea's requirements;
- in the spirit of open development - if I can release code I'm not proud of, then I am less hypocritical encouraging others to do the same;
- in case anyone wants to improve it, since it is still in active use in Swansea.
- Takes in a list of students, who are divided into 'pairs' and 'cohorts'
- Takes in four lists of barcodes for assessment points throughout the year
- Uses a Monte Carlo/Simulated Annealing-style algorithm to assign activities to assign a range of available activities to pairs, with certain constraints. Unfortunately these are hard-coded in one way or another
- Attempts to optimise the schedule to minimise "unpleasantness"
- Generates personalised timetable cover sheets for each student, including their name, pair number, cohort, experiment listings, and barcodes for assessment points.
- Generates a summary page of how many of each experiment are due to occur on any given week, allowing the lab techs to avoid wasting time setting up unnecessary kit.
- Generate mark sheets allowing markers to easily identify students whose work they need to mark.
- Identifies students' preferred language and generates their timetable using this language. This is done in a fairly extensible way, although not using any standard internationalisation libraries.
Set up a virtualenv running Python 3.5+ (older versions will probably work,
but I've not tested them). Then use pip install -r requirements.txt to
install the prerequisites.
To generate cover sheets with the Swansea University branding, you need to have appropriate fonts in the same directory as the code. I can't stick these on GitHub, so you'll have to find them yourself. The files you'll need are:
Cosmos-Light.ttfFutura-Book.ttfFutura-Heavy.ttf
There's a small FontForge script in the repository that will convert .otf fonts to .ttf, since I only had these fonts in .otf format. Alternatively you can comment out the font load lines and replace all other font references so that it uses a font you do have instead.
You will need five or six files in the repository directory:
students.csv: A CSV file containing columns: index, student ID, surname, first name, pair number, language code (this format can be obtained by massaging the result of copying/pasting the module listing from the SU Intranet into Excel; in principle it could be scripted); and/orpairs.csv: a CSV file in the above format but anonymised to have pair number instead of student number, and blank entries for namebarcodesAB.csv,barcodes1.csv,barcodesCD.csv,barcodes2.csv: tab-delimited files containing two columns: student number, and barcode, for the assessment points mid-semester 1, end of semester 1, mid-semester 2, and end of semester 2, respectively. (This is the format generated by the SU College of Science Intranet.) Note that students on Welsh-coded modules have separate assignments so will need to get added manually.
Some parameters are hard coded, including the number of pairs and the split of pairs into cohorts. The experiment list is also hard coded. Some of these need to be tuned manually each year; see below.
To generate a timetable, run python generate_timetable.py new. This can
be done even before you have a student list to generate against, although
the cover sheet generation will fail. This is useful because it sometimes
takes a few attempts and some massaging of parameters to get the algorithm
to converge on a real solution.
It probably won't be able to converge to zero unpleasantness, so once it seems to be plateauing then hit Ctrl+C to abort and output the current best configuration. (See the source for more about badness and unpleasantness.)
(I have been known to on occasion manually read in the database at this point and switch some experiments around to reduce the unpleasantness in a way the algorithm couldn't spot, although I've since fixed the bug that caused that.)
Once the student list is available, then a fresh set of cover sheets can
be generated using python generate_timetable.py. The anonymised versions
are obtained using python generate_timetable.py anon.
The experiment count listing for the technical staff can be generated using
python print_week_list.py.
Mark sheets are generated for the entire year by running
python print_mark_sheet.py 1+2. The 1+2 can be changed to only
print the mark sheet for a single semester if so desired; this seems popular
with markers so should probably be the standard going forward.
There are also some ancillary files from when I have needed to print out a set of barcode stickers (because the barcode export wasn't ready in time for the initial printing), and print marking lists for an individual week for the whole group (rather than a single marker for a year/semester). These are not up to date, but might be useful in some circumstances.
Because this was written in a hurry, some many things require manual tuning.
- Number of pairs to generate (fewer makes for easier scheduling)
(in
lab_mc.py) - Division of pairs into cohorts (in
lab_mc.py) - The date of the first Thursday of each term (i.e. the first lab
day after the summer, after Christmas, and after Easter) (in
weeks.py)