You will need a Python installation. The code was developed on Python 3.9 but should run on any newer Python3 version. You can install dependencies manually from requirements.txt or with pip install -e ..
You can then call the script to solve the matching problem with python ./ayto/ayto.py --yaml_file_path <YAML_FILE_PATH>, where <YAML_FILE_PATH> is a path to a yaml file which decodes the names of all people, matching night couples and results as well as truth booth decisions. An example is given in ./examples/AYTO_Season2_Germany_AfterEp18.yaml.
This code is untested and rather a PoC than production ready code. It was developed for the German seasons of AYTO airing on RTL. It silently assumes that we have "Males" and "Females" and we only match a Male with a Female. Furthermore, all matches and truth booth couples need to be defined in the yaml file, first with the male name followed by the female name. This was done because of pure laziness and should in no way be meant to discriminate anyone. Code modifications would be needed to enable a season like "Come one, Come all" (season 8 AYTO US), where anyone can match with anyone. This also makes the optimization problem more complex and, hence, harder to solve.
We use a MIP solver to solve the problem. We solve min norm(x,1) s.t. Ax=b, x in {0,1}where we actually already now the minimum and it is not unique until we have enough constraints (i.e. matching nights and truth booth results). This is derived from a compressed sensing approach where we would solve min norm(x,0) s.t. Ax=b since we know that our solution is sparse: The vector x of length males * females decoding all possible matches is s-sparse where s defines the number of couples. This knowledge allows us to solve the underdetermined linear equation system Ax=b nevertheless. We use a mixed integer linear programming approach (MIP). This algorithm will in our case always find a feasible and optimial solution, that means we always find a valid combination of couples that satisfies all information we have from matching nights and truth boothes. However, we do not know and do not check if this solution is unique. This approach is derived from a compressed sensing approach using linear optimization trying to do sparse signal recovery (see [https://en.wikipedia.org/wiki/Compressed_sensing]) and not treating this as a combinatorical problem. For the latter see e.g. [https://blogs.sas.com/content/operations/2018/08/14/are-you-the-one/] for an example.