1
$\begingroup$

We have distributed data centers and we build decision trees in each data center. Our problem is to combine our CART decision trees into one CART decision tree. The data in each data center related to the same event (data from light sensor for instance). I know about boosting methods but they don't give result we want to.

Is there any known method to do this ?

$\endgroup$
  • $\begingroup$ Do you really want to merge them into a single tree (why?), or would a "forest" consisting of several trees that get averaged also be acceptable? $\endgroup$ – stmax Oct 20 '16 at 14:47
3
$\begingroup$

You mention two decision trees. Traversing a decision tree is very cheap, so running a feature instance through multiple trees is very fast, you could just take all the decision trees from the data centers and average the outcome, maybe weigh it by the (crossvalidated) strength of the models. Random Forests are powerful models that also combine decision trees in this way, except that they are done on random subsets of the features (and in some cases also random subsets of the data).

|improve this answer|||||
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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