I have a set of hierarchical classes (ex. "object/architecture/building/residential building/house/farmhouse"), and I build a tree where each node is a classifier. However, the appropriate class for a particular featureset could be on any level (e.g. "object/architecture/building").

Currently, I use two different methods-->

  1. I make locally optimal decisions unless the predicted class is a leaf node, where I divide its probability by a constant (hyperparameter), and then recheck to get the class with the highest probability until I hit a leaf node.
  2. At every node, I trace through the 3 children with the highest probabilities, and then propagate the probabilities down to the leaves (by just summing the logs). Then I add a hyperparameter based on the level down the tree (the deeper the level, the higher the score).

Both of these methods were implemented to account for the system most often deciding that a class should be too high up in a tree (that is, in a decision between say object/architecture/building/house and object/architecture/building, without the constants, the system will almost always prefer object/architecture/building).

This is not good--I realize I should not have this constant in the mix, but I'm not sure how to accomplish the task without it. Also, if I could somehow not have to use probabilities (and could use just distance to hyperplanes instead (so I could use svms instead of logistic regression as the classifiers in each node)), that'd be ideal. (I'm using python and scikit learn, just to be specific)

Any thoughts?


1 Answer 1


Here is one standard solution, create a new tree from the hierarchy and add a leaf-node "other" under every non-leaf node. This "other" node contains all the positive example which do not fall in any of the child-nodes.

For example, if you are hierarchy is

{animals : {sea:{fish,shark}} , {land:{lion,elephant}} }

you're new hierarchy will look like

{1 : {2:{fish,shark,sea(other)}} , {3:{lion,elephant,land(other)} , animals(other) }

Hope that makes sense.

  • $\begingroup$ Hi Siddharth--I actually do exactly that--I should have specified that in the original post--my bad. It definitely helps, but does not remove the need for the hyperparameter to push further down the tree. $\endgroup$ Jul 30, 2013 at 0:31
  • $\begingroup$ (that is, unless I introduce a hyperparameter to make the xx(other) less desirable, the system will choose it too early) $\endgroup$ Jul 30, 2013 at 17:28
  • $\begingroup$ I'm not sure I fully understand, so let me rephrase what I said earlier - If you restructure the tree, then there are no examples that belong to a non-leaf node, you stop when you hit the leaf node. Edit: Since you have multiple labels, one solution is to keep going down on all paths where the probability of selecting the node > .5 $\endgroup$ Jul 30, 2013 at 17:44
  • $\begingroup$ That's true--there are no examples that belong to a non-leaf node, and stop when you hit the leaf node. In your example though--> {1 : {2:{fish,shark,sea(other)}} , {3:{lion,elephant,land(other)} , animals(other)} , my system will almost always want to stop at animals(other) without a specific penalty for stopping there. With the penalty introduced, I get around ~73% accuracy, although it is still lower than ~76% for the independent/flat case (i.e. instead of a tree, all the classes were considered independently with one classifier). $\endgroup$ Jul 30, 2013 at 18:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.