Student and instructor facing website that allow students to provide availability and instructors to divide students into groups based on availability.
- Students get to advertise their availability in their own timezone. No need for timezone conversions.
- Instructors can manually group students based on a grid layout, or use an automated scheduler with a configurable group size that seeks to maximize the number of hours all members of the group are available.
| Empty | Filled Out |
|---|---|
 |
 |
| Empty | Filled Out |
|---|---|
 |
 |
The site is live at https://groups-server.shuttleapp.rs/student for students and https://groups-server.shuttleapp.rs/instructor for instructors.
Questions or issues can be resolved by clicking on the issues tab above an opening a new issue, or by sending me an email at kunzep@byui.edu.
Run build_and_test_site_local.sh in project root. A local webserver will be spun up and the IP address + port will be printed to the console.
Run build_and_deploy_site.sh in project root.
BASE_URL is printed as a result of deploying or running the local webserver.
- BASE_URL/student. Students fill out the form there and send the instructor the schedule code they receive, either through email, a survey, etc.
- BASE_URL/instructor. Instructors paste in all of the student schedule codes they receive. They can use the interface to manually split into groups based on availability, or have it done automatically.
- BASE_URL/random (local webserver only). Get 50 random schedule codes for testing instructor functionality.
The core group scheduling code is written in Rust and runs in the browser after being compiled to WebAssembly. This code also handles encoding and decoding schedule ids (base64 encoded strings that compactly encode student information and a bitvector of student scheduling information). The group scheduler has a plugable architecture that allows it to use different group assignment algorithm. Originally it used a hill-climbing algorithm with random re-starts to avoid getting stuck in a local minima. It created a random group assignment, then randomly swaps students as long as a swap results in a better objective function for the entire group assignment. Now it uses a hill-climbing algorithm from this paper which has a better way of ranking the goodness of team assignment and seeks to maximize the minimum team score in an assignment.
I considered other search algorithms (simulated annealing, genetic search, etc) and constraint solvers (this problem's formulation is similar to the wedding seating problem) but the main barrier lies in implementing a better objective function. This function should maximize the number of hours (especially consecutive hours) each team members in a group have in common, while attempting to make all groups equally good (we don't want some very good groups that maximize the objective function but that overshadow some very bad groups). It's possible the Gini coefficient is how we could approach this. With a better objective function, we could use a more sophisticated search algorithm to attempt to maximize it. As it is, the current hill-climbing methodology finds the best possible group assignment relatively quickly, as shown by plotting the convergence in unit tests with random data. Real student data is not random so it remains to be seen how this will perform in the real world.
https://info.catme.org/features/team-maker/ also has some interesting ideas.