How to extend ensemble methods in scikit-learn with a new learning algorithm

Question

I have a new algorithm for the decision tree ensemble regression method that I need to implement, and I would like to build on the infrastructure that a Python-based scikit-learning package provides, if possible. I want to create a new type of base decision tree and a new way to combine them to create ensemble regression forecasts.

The documentation for scikit-learn is for use, not extension. I looked through the code of the existing ensemble methods, but there is a pretty subtle Python-fu that makes it difficult to figure out where to start. Does anyone know the documentation that explains how to extend scikit with custom classes? Or is there an example of a simple extension of this kind? (It can be on any module, it should not be ensembles.)

Thank.