Fixing sparse matching

[mortonjt]

This fixes the issue with sparse matching with #290 . We still need to fix travis to get this in.

[mortonjt]

I've also updated both matching methods so that the set ops don't add any additional randomness.
Turns out that set ops are inherently unstable, no random seeds can make the set ordering reproducible.
So after taking intersections, these sets are converted to lists and sorted.