使用2个分类器进行集成学习

Question

I'm trying to combine 2 approaches to classifying my data, one comes from a SVM and another external classifier that gives out one or more labels as to what it thinks the observation point is. Is it possible to get these two classifiers to work together? Similar to what boosting does.

I've noticed that in scikit's adaboost implementation, it takes in only 1 type of classifier. Further, the 2nd classifier that I have, which gives out 1 or more labels doesn't have any "weights" associated with it. How do I go about this?

Answer 1

There is a technique called stacked generalization, which basically takes the output of any K classifiers and trains a second layer classifier on top of this - so the input to the second layer classifier is the output of the first layer classifier. You can either use crossvalidated and averaged label predictions, or predict_proba outputs, or some other metric.

A nice introductory link on this is here