ProjectAllocation

The purpose of this algorithm is to assign students to projects based on their preferences. The allocation is done in such a way that it seeks to maximize the number of students whose preferred choices are fulfilled. This follows the principle of utilitarianism, developed by UCL founder Jeremy Bentham, which states that "the best allocation of projects is the one that ensures the greatest good for the most students."

Getting started

Create 2 csv files or amend those included in the repository with actual data

• student_preferences.csv with 1 + number of preferences columns: student, preference1, preference2, ..., preferenceN
• project_capacities.csv with 3 columns: project, capacity, supervisor

Run the allocation algorithm in matching.ipynb file.

The output are two csv files with the results after the alogrithm is run once.

• matching_student.csv for a student focused view with 4 columns: student, project allocated, preference rank of allocation, list of preferences
• matching_project.csv for a project focused view with 4 columns: project, list of student allocated, capacity, supervisor

Random seed searching

Run the random seed searching to maximise student happiness (with parallel processing) using python seed_search_par.py. This script runs the process 16.000 times, testing different orders of students as proposers of the matching, and returning the one match, with the start seed, that maximises happiness 😊, across all those iterations.

Allocation Algorithm

The allocation is conducted using the Gale-Shapley Algorithm, where students are the "proposers" and the projects are the "acceptors". This setup will be student-optimal, meaning students will generally have their most preferred choice and no student allocated to a more desired project would have a lower preference than you.

For example project 1 with one capacity, it is student 1's 1st choice, student 2's 2nd choice, and student 11's 1st choice. Only students 1 or 11 will be allocated to project 1.

In order for all students to have equal opportunity to be allocated to their desired project and have enough time to speak with potential supervisors, students are uniformly randomly selected (i.e. the preferences of projects to select students are random).

The quality of the matching is measured using the rate_matching function. This calculates a score per student and their preference, the overall score of the matching is

$$ S = 10\sum_{i=1}^N r_i^{-2} $$

where $r_i$ is rank preference of a student (e.g. 1 is top preference, 3 is third preference), N is the number of students, 10 is an arbitrary scaling factor for readability. We will run the algorithm multiple times to choose the best matching that maximises student happiness.

Limitations

For a complete matching to be guaranteed, all students would have to provide preferences for all projects. However, this is impractical. Thus in particular circumstances, such as when too many students have the exact same preferences, it is possible that some students are not allocated to any of their preferences. In these cases, allocations will be manually conducted to the remaining projects.

Example

Example student preferences and project capacities have been provided to demonstrate the allocation algorithm in action.

FAQ