see http://scikit-learn.org/dev/auto_examples/ensemble/plot_forest_iris.html right-most column. The decision boundary does not take the weights into account - these should be used instead of `alpha=0.1`. reported by Ian Ozsvald on the sklearn mailing list.