This is an unofficial Python implementation of the Nobel Prize winning algorithm used to match Canadian medical school graduates to residency positions.
This is largely standalone, the only requirement being Python 3 and pandas to manage the csv rank order lists.
pip install pandas
Tests can be run using python -m unittest in the root directory.
The test case here is based on the Carms guide to the matching algorithm. Note that the example used in the test case is from a prior version of this page. Print statement have been included as comments if you wish to run a "loud" version of the matching in order to show the search and recursive replacements as they happen.
This implements the simple markets, applicant proposing version of the algorithm. The main differences from the full carms use is the lack of couples matching or second year entry.
The algorithm is based on the work of Alvin E. Roth and Elliot Peranson in their paper The Redesign of the Matching Market for American Physicians: Some Engineering Aspects of Economic Design as well as previous work on matching markets by David Gale and Lloyd Shapley. Roth and Shapley jointly won the 2012 Nobel Prize in Economics "for the theory of stable allocations and the practice of market design"
The goal is to form "Stable Matches", that is, no candidate and program wish they were matched with each other instead of their actual match.
To do this both candidates and programs produce rank order lists of their preferences after which the algorithm progresses through each candidate, down their rank order list, temporarily matching them to their most preferred program which has also ranked them.
As students are considered, better matches, from the perspective of the program, may displace candidates with a temporary match in which case that candidate progresses down their remaining options potentially displacing others and so on until each students rank order list has been exhausted.
At this point all students will either be matched to the best program they can be, relative to their competition, or will have failed to match entirely.
The design of the algorithm minimises the potential for strategic rank order lists, incentivising programs and applicants to state their true preferences.
Further work needs to be done on this implementation to support candidates with complementarities.
- John Dirk Morrison - Research and Implementation - Github: J-DM
You can reach me at john.dirk.morrison [you know the thing] gmail.com