Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Any good implementation of greedy set cover for large datasets? [closed]

This question follows from a related question of mine posted here. @mhum suggested that my problem falls into the covering problem domain. I tried encoding my question into a minimum set cover problem and currently I have a dataset in this form:

Set        Cost
(1,2)        1
(1)          1
(1,2,3)      2
(1)          2
(3,4)        2
(4)          3
(1,2)        3
(3,4)        4
(1,2,3,4)    4

The objective is to find a good set cover that covers all numbers and one that attempts to minimize the total cost. My dataset is big with at least 30000 sets (of size varying from 5-40 elements) like this. Are there any good greedy implementations to solve this or am I on my own to implement this? I am not an expert in LP but any LP-solvers (from numpy/scipy) that can solve this are also acceptable.

like image 324
Legend Avatar asked Oct 29 '11 00:10

Legend


2 Answers

There is a well-known greedy approximation algorithm for set cover that is also easy to implement in whatever language of your choice. The algorithm itself is described here:

http://en.wikipedia.org/wiki/Set_cover_problem#Greedy_algorithm

It is so simple that the easiest thing is just to write it from scratch.

Notably, it is also the best polynomial-time approximation algorithm known for set cover. That means that to get better worst-case performance (more compact result set) you would need to have non-polynomial running times (= slow algorithms for large sets).

Unfortunately the Wikipedia entry doesn't actually cover weighted set cover, which is the case here. The extension is simple, and is described e.g. here:

http://pages.cs.wisc.edu/~shuchi/courses/880-S07/scribe-notes/lecture03.pdf

Some more useful notes:

http://www.cs.ucr.edu/~neal/non_arxiv/Young08SetCover.pdf http://www.cs.uiuc.edu/class/sp08/cs473/Lectures/lec20.pdf

like image 120
Antti Huima Avatar answered Nov 15 '22 18:11

Antti Huima


My linear time / space implementation of greedy set cover in c++ is available at github.

https://github.com/martin-steinegger/setcover

A calculation for 40.000.000 sets with avg. 10 elements per set takes around 4 Minutes on computed on Amazon AWS m2.2xlarge instances.

I still work on some tricks to improve the performance

  1. remove subsets that are covered by a bigger set with MinHash
  2. remove all sets that just contain one element that is no other set
like image 25
martin s Avatar answered Nov 15 '22 16:11

martin s