Thursday, February 24, 2011

Improving Branching. Part. II: Polarity

In previous post I concluded that variable's polarity searched first affects the solving time of some instances (mostly of satisfiable instances of course). With this in mind I ran four tests with the branching heuristic modified as follows

  • initial - branch first on positive polarity
  • plus - branch first on polarity with the highest score
  • minus - branch first on polarity with the lowest score
  • random - branch first on a random polarity
The results are in the graph below. Needless to say that random rocks. plus and minus performed similarly. By choosing a random polarity STRUCTure can now solve three more instances from the AProVE class (I am developing negative emotional feelings towards them).

No comments:

Post a Comment