Discover MakerZone

MATLAB and Simulink resources for Arduino, LEGO, and Raspberry Pi

Learn more

Discover what MATLAB® can do for your career.

Opportunities for recent engineering grads.

Apply Today

Viewing the tree used for prediction by the method of "RUSBoost' in fitensemble

Asked by uwoldtimer on 18 Jan 2013

I have a question in regard to viewing the Tree from the fitensemble function. I am using 'RUSBoost' as the method. I can see that there are 1000 trees in the cell called Trained since I set nlearn to be a 1000. So these 1000 trees are the weak learners if I am not mistaken. But then where is the strong learner that was gotten using these weak learners? In other words, where is the tree that is actually used for prediction? How can I see that tree?

Thank you for your help.

0 Comments

uwoldtimer

Products

No products are associated with this question.

1 Answer

Answer by Ilya on 18 Jan 2013
Accepted answer

The strong learner is the ensemble. An ensemble is a collection of trees. It predicts by averaging predictions from individual trees. This average is weighted. You can get the weights from the TrainedWeights property of the ensemble object.

4 Comments

Ilya on 19 Jan 2013

If an ensemble could be reduced to a single decision tree with univariate splits, there would be no point in growing the ensemble. If you are willing to consider non-univariate splits such as, for instance, splits on various functions of two or more predictor variables, you can use that flexibility to represent any complex decision boundary in a multivariate domain (consider gene expression programming, for example). Ensemble learning does not do that. An ensemble is a non-parametric model, and here is the usual feature of non-parametric models - it is hard to visualize them in the multivariate space.

uwoldtimer on 30 Jan 2013

I have another question related to this. Among all the machine learning algorithms provided in the MATLAB toolboxes, is fitensemble the only algorithm that allows the input of a cost function?

Ilya on 31 Jan 2013

Please open new threads for new questions.

ClassificationDiscriminant, ClassificationTree and ClassificationKNN accept the cost matrix as well.

Ilya

Contact us