Grouping students across subject areas without predeterming the instructor

[MrGardiner]

Hi,

We have courses like Math which has multiple subparts for different meeting times, e.g.
MATH
Lec (1x1hr) Instructor A
-Lec (1x80mins) Instructor A
--Lec (1x40mins) Instructor A

and multiple parts like this (many of these MATH courses):
Lec (1x1hr) Instructor B
-Lec (1x80mins) Instructor B
--Lec (1x40mins) Instructor B

My understanding is that a student who enrols in the first subpart must also then be in all the children classes (which is good - I want the same group of students in all 3 subparts/will have the same instructor in this case). Then another course like English which might be similar:
ENGL
Lec (1x1hr) Instructor C
-Lec (1x40mins) Instructor C
Lec (1x1hr) Instructor D
-Lec (1x40mins) Instructor D

I want to further group students together, so I want students who do the MATH course together to also be in the ENGL course together. I have determined these students in advance. I think with reservations I can create 'Group 1', and reserve space for 'Group 1' in the first Math course and the first ENGL course. However, the issue is that by doing this I'm now defining who their instructors are, which I think has significant impacts on being able to construct a viable timetable. I've considered leaving instructors for these courses blank and assigning them via instructor scheduling/teaching preferences, however my understanding is that the rooms are allocated in the first step (course solver), so the rooms can't be shuffled around to match instructor needs at this point.
The other way I thought to group them was to have the ENGL subparts children of the MATH ones, but I don't think that helps either (potentially worse).

Is there a way to group students across courses, without predeterming who the instructor is?

[tomas-muller]

Firstly, a student needs to have one class for each scheduling subpart of a single configuration, following the parent-child relations when defined. So, for instance, if there is one Lecture subpart with two lectures (Lec 1 and Lec 2) and one Lab subpart with eight labs (Lab 1, 2, ... 8) that are nested (Lab 1 - 4 are under Lec 1 and Lab 5 - 8 are under Lec 2), the selection of a lab for a student dictates in which lecture the student must go as the parent class must be taken. So, students in Lab 1 - 4 must have Lec 1, and students in Lab 5 - 8 must have Lec 2. However, every student will only have one lecture and one lab class.

Secondly, the course timetabling solver in UniTime was designed to handle individual students and their course demands, assign each student to their classes, and minimize their student conflicts individually. It does not assign groups of students (usually referred to as cohorts) to classes. There are a few ways around that, however.

First of all, the student sectioning algorithm that the course timetabling uses can try to keep students from the same cohort together. This is usually used when there are curricula (and curricula course requests are used, so students of the same curriculum and classification are preferred to be kept in the same classes). Still, it can also be used for student groups when student course requests are used, and the student groups are of a student group type with the Keep Students Together toggle checked. You may also need to switch the Student Sectioning parameter in the solver configuration to Branch & Bound or Local Search algorithm. The B&B Group option can also be interesting as it tries to keep students together first, and only minimizes student conflicts second. Enabling On Fly Student Sectioning may also help if you are using UniTime 4.9 or UniTime 4.8.170 or later, as the solver will try to swap students around during the search, not only initially and afterwards.

The second option is to use reservations (typically curriculum or student group reservations) to assign students in a particular cohort to a specific class. That usually works quite well, unless the individual classes have very restrictive requirements (e.g., each instructor has only one or two available times when they can teach their class).

The last option is not to care and let the solver do its job while working with individual students. The initial sectioning (assignment of students to individual classes based on their course demand) already tries to keep students with similar courses together, and the solver does a pretty good job of keeping the number of student conflicts low. It can be further adjusted in the solver configuration. Besides the options already mentioned, it is also possible to increase the student conflict weights, or to make student conflicts hard (not allow the solver to have any, or only when there is no other way).

By default, student conflicts are considered soft constraints. This means that student conflicts are allowed, but the solver tries to minimize their number (together with other criteria like time and room preferences). The following solver parameters allow the solver to consider certain or all student conflicts as hard constraints (cannot be violated).

• Joint Enrollment Constraint: conflict limit (% limit of the smaller class), defaults to 1.0
• Joint Enrollment Constraint: limit weakening, defaults to 0.001

For instance, if the first parameter is set to 0.1, two classes cannot overlap if they share more than 10% of students of the smaller class. You can even set it to 0.0, not allowing any student conflicts at all, but the solver may not be able to find a complete solution after that.

The second parameter (named "Joint Enrollment Constraint: limit weakening") allows this constraint to weaken as the solver runs if it is struggling with the constraint (the first parameter is increased by this number every time the constraint causes a conflict). This is useful when you want to put more pressure on the student conflicts without losing the ability to find a complete solution when there are a few cases where this constraint cannot be satisfied. When you set it to 0.0, student conflicts will stay hard (but the solver may be unable to assign all classes).