2013/7/7 Ian Ozsvald <[email protected]>: > Following on from the previous post, I thought (from reading only and > accepting no prior experience with AdaBoost) that the main goal of > AdaBoost was to combine weak classifiers (e.g. a depth-restricted > DecisionTree) rather than building an ensemble of strong classifiers > (as in e.g. a RandomForest). > > The example on the site: > http://scikit-learn.org/dev/auto_examples/ensemble/plot_forest_iris.html > uses DecisionTrees with max_depth=None for each of the 4 classifiers. > Using a depth restricted classifier (e.g. max_depth=3) for AdaBoost > results in the same classification quality in this example. > > Might the example say more about AdaBoost's ability to use weak > classifiers if we used a restricted depth DecisionTree?
+1, PR accepted :) Boosting is good for ensembling a large number of underfitting models and thus correcting their individual bias. Bagging and other randomized voting aggregates is good for ensembling a large number of overfitting models and thus correcting their individual variance. -- Olivier http://twitter.com/ogrisel - http://github.com/ogrisel ------------------------------------------------------------------------------ This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
