Given two sets of objects, in some kind of container indexed by small integers, and a metric function which computes distances between objects in one set and objects in another, then this code computes the mapping between the two sets which minimises the total distance.
For a long time I thought this must be equivalent to something like the travelling salesman problem, but it's actually much more tractable.
This contains a Python package, installable with 2.6 and 2.7 at least,
which implements the algorithm. There are unit tests, a setup.py
and so on. You need NumPy to use it, and
nose to run the unit tests.
There is documentation for this.
This is the original proof-of-concept, implemented in Racket. It has no pretensions to be production-quality code: it's just here for the record.
I have removed the Travis CI tests: it should still pass any tests
that it used to as nothing else has changed: make test in the
python directory will run them.
There are unit tests for the Racket module, which you can run with
raco test if you want (or make test in the racket directory).