A Stochastic Local Search Approach for Learning Chordal Markov Networks

Developed by the Constraint Reasoning and Optimization Group
at the Department of Computer Science, University of Helsinki.

The program finds a chordal Markov networks given a CSV file.
Details on the algorithm can be found in [1].


Usage: ./lsmarkov  [arguments...]

Arguments (defaults in parentheses):
  --greedy                Use greedy search instead of a stochastic one. (false)
  --restart-rate <value>  Maximum number of iterations without local improvement. (1000)
  --ess <value>           Sets the equivalent sample size when computing scores. (1)
  --max-clique <value>    Limits the maximum clique size (for CSV inputs). -1=limitless (-1)
  --precompute <value>    Precompute clique scores (for CSV inputs) 0=no, 1=some, 2=all (0)
  --dynamictw             Start with small clique limit and then increase (3, 4, 8, 16, ...) (false)
  --fixed-iters           Have a fixed number of iterations between restarts. (false)
  --chow-liu              Use Chow-Liu to find the optimal tree in the beginning. (false)
  --operations <value>    Operations to use; [C]lique, [E]dge, [V]ertex; e.g. CE ("CEV")
  --seed <value>          Set a custom RNG seed. 0=random (0)


The source code of the C++ implementation can be downloaded here.


[1] Learning Chordal Markov Networks via Stochastic Local Search
Kari Rantanen, Antti Hyttinen and Matti Järvisalo.
In Proceedings of the 24th European Conference on Artificial Intelligence (ECAI 2020),
pages ???—???.