Tree ensmbles now have a new decision_path method that rturns a boolean matrix indicating which item falls under which node. At the same time the existing transform method on these estimators is being deprecated.

The decision path is a very usefu set ofl fetures and would be a ntural transform for these estimators,
once the original transform is removed.
In particular it would make it very simple to implement Jerome Friedman's Imprtance Sampled Learning Ensembles.