Implement errorTree in DiscreteFactor

[varunagrawal]

Common base class method to reduce duplication.

[dellaert]

Hmmmm. I'm worried about complexity here, especially if we have pruned branches. I think you replaced a sometimes errorTree with an always expensive version.

[varunagrawal]

Hmmmm. I'm worried about complexity here, especially if we have pruned branches. I think you replaced a sometimes errorTree with an always expensive version.

Interesting. I copy-pasted the code from DecisionTreeFactor to DiscreteFactor since TableFactor was anyway calling the DecisionTreeFactor version.
I didn't see this potential issue because except for discreteKeys there is no class specific method being used here. And discreteKeys should be the same for both DiscreteFactor and DecisionTreeFactor.

Maybe the original implementation was already expensive to begin with?

[dellaert]

Actually, you are right! We should just call the DecisionTree constructor in the DTF. That is a potential speedup! So, I like your base class implementation for generic, but it should speed up for derived class. Probably also in TableFactor: is toDecisionTreeFactor efficient?

[varunagrawal]

Probably also in TableFactor: is toDecisionTreeFactor efficient?

It should be. It's from @ywkim0606's original PR which you had worked on with him.