corail-research/learning-generic-csp

How could we "condense extension constraints

Closed this issue · 0 comments

Following tests with this first implementation, I found -as expected- that extension constraints make the graph explode. Moreover, I also found that many of them could be condensed. For example, say we have the following tuples:

  • (1, 2)
  • (1, 3)
  • (1, 4)
  • (1, 5)
  • (1, 6)
  • (1, 7)
  • (1, 8)
  • (1, 9)
  • (1, 10)

When we think about this constraint, we think that the first variable can take the value 1 and the second variable can take an integer value in the range {2, 3,..., 10}. Using a machine-readable analogous version would make the graphs a lot smaller, which in turn would make learning easier.