Why?

Sometimes, specially during algorithm competitions, a Union-Find data structure is needed to solve a certain problem (e.g. when implementing Kruskal’s MST algorithm) In those cases, we usually use whatever off-the-shelf implementation we happen to find in PyPI and, many times, we get the infamous submission fail Time Limit Exceed(TLE) because that specific Disjoint-Set implementation is not fast enough…

Worry not! A fast and simple enough (you can copy paste the code here) disjoint-set implementation is now available:

Installation

$ pip install dset

Usage

>>> import dset
>>> s1 = dset.Set()                 # To create new sets objects (it runs in O(1))
>>> s2 = dset.Set()
>>> s3 = dset.Set()
>>> dset.groups()                   # To check the number of different groups (it runs in O(1))
3
>>> dset.find(s1) == dset.find(s2)  # To check if two sets are in the same group
False
>>> dset.union(s1, s2)              # to group s1 and s2 sets together
>>> dset.groups()                   # Now there are only 2 groups: s1-s2 and s3
2
>>> dset.find(s1) == dset.find(s2)  # They are in the same group s1-s2
True
>>> dset.find(s2) == dset.find(s3)  # They are in distinct groups s1-s2 and s3
False
>>> dset.union(s2, s1)              # do nothing (no need to check if s1 and s2 belong to the same group)
>>> dset.groups()                   # Same as before only 2: s1-s2 and s3
2
>>> dset.union(s1, s3)              # group s1-s2 and s3 together.
>>> dset.groups()                   # Now there is only one group: s1-s2-s3
1
>>> dset.find(s1) == dset.find(s2)  # All the sets are in the same groups s1-s2-s3
... dset.find(s2) == dset.find(s3)
... dset.find(s1) == dset.find(s3)
True
True
True

Future improvements

• Make official Sphinx Documentation. (minor releases)

• Make test suite. (minor releases)

• Change to a list-based implementation (major-minor release)

  • Improve running time (constant terms)

  • Improve memory space (constant terms)

• Make the package C-based. (major-minor release)

  • Improve running time (constant terms)

Notes on Complexity Running Times

Separately, either “Union by Rank” or “Path Compression” improves the running time of the operations on disjoint-sets. Alone, “Union by Rank” yields a running time of O(m log n), where m is the number of Union and Find-Set operations and n is the number of sets.

The improvement is even greater when we use the two heuristics together; the worst-case running time is O(m f(n)), where f(n) is a very slowly growing function, the Inverse Ackermann function (e.g. for n equal to the number atoms in the universe, f(n) is only 4) So for any conceivable application, we can consider the running time of m Union + Find-Set operations to be O(m) and therefore both Union and Find-Set operations have O(1) amortized running time in practice.