For trees γ parameterizes the split variables and split points at the internal

# For trees γ parameterizes the split variables and

• Notes
• 31

This preview shows page 15 - 25 out of 31 pages.

For trees, γ parameterizes the split variables and split points at the internal nodes, and the predictions at the terminal nodes [2] c 2019 The Trustees of the Stevens Institute of Technology

Subscribe to view the full document.

Bagging and Bumping Bagging and Random Forests Boosting Methods Boosting Trees Fitting Typically these models are fit by minimizing a loss function averaged over the training data: min { β m m } M 1 N X i = 1 L y i , M X m = 1 β m b ( x i ; γ m ) ! Or a simple alternative being fitting a single basis function: min β,γ N X i = 1 L ( y i , β b ( x i ; γ )) [2] c 2019 The Trustees of the Stevens Institute of Technology
Bagging and Bumping Bagging and Random Forests Boosting Methods Boosting Trees Algorithm 10.2: Forward Stagewise Additive Modeling 1. Initialize f 0 ( x ) = 0 2. For m = 1 to M 2.1 Compute ( β m , γ m ) = arg min β,γ N X i = 1 L ( y i , f m - 1 ( x i ) + β b ( x i ; γ )) 2.2 Set f m ( x ) = f m - 1 ( x ) + β m b ( x ; γ m ) [2] c 2019 The Trustees of the Stevens Institute of Technology

Subscribe to view the full document.

Bagging and Bumping Bagging and Random Forests Boosting Methods Boosting Trees Squared Error Loss L ( y , f ( x )) = ( y - f ( x )) 2 we have: L ( y i ; f m - 1 ( x i ) + β b ( x i ; γ )) = ( y i - f m - 1 ( x i ) - β b ( x i ; γ )) 2 = ( r im - β b ( x i ; γ )) 2 where r im = y i - f m - 1 ( x i ) is the residual of the current model on the i th observation c 2019 The Trustees of the Stevens Institute of Technology
Bagging and Bumping Bagging and Random Forests Boosting Methods Boosting Trees Exponential Loss L ( y , f ( x )) = e - yf ( x ) Using this results in AdaBoost.M1 being equivalent to forward stagewise additive modeling. (as will be shown) c 2019 The Trustees of the Stevens Institute of Technology

Subscribe to view the full document.

Bagging and Bumping Bagging and Random Forests Boosting Methods Boosting Trees We use the basis functions as the individual classifiers, and so using the exponential loss function, we have to solve: ( β m , G m ) = arg min β, G N X i = 1 e - y i ( f m - 1 ( x i )+ β G ( x i )) for the classifier G m and the coefficient β m c 2019 The Trustees of the Stevens Institute of Technology
Bagging and Bumping Bagging and Random Forests Boosting Methods Boosting Trees We can express this as: ( β m , G m ) = arg min β, G N X i = 1 w ( m ) i e - β y i G ( x i ) with w ( m ) i = e - y i f m - 1 ( x i ) . This solution can be obtained in two steps. First for β > 0 we have: G m = arg min G N X i = 1 w ( m ) i I { y i 6 = G ( x i ) } c 2019 The Trustees of the Stevens Institute of Technology

Subscribe to view the full document.

Bagging and Bumping Bagging and Random Forests Boosting Methods Boosting Trees Our criterion in ( β m , G m ) becomes: e - β X y i = G ( x i ) w ( m ) i + e β X y i 6 = G ( x i ) w ( m ) i which can be written as: ( e β - e - β ) N X i = 1 w ( m ) i I { y i 6 = G ( x i ) } + e - β N X i = 1 w ( m ) i c 2019 The Trustees of the Stevens Institute of Technology
Bagging and Bumping Bagging and Random Forests Boosting Methods Boosting Trees If we plug this G m into the earlier equation, we have: β m = 1 2 log 1 - err m err m where err m = N i = 1 w ( m ) i I { y i 6 = G m ( x ) } N i = 1 w ( m ) i The approximation is then updated f m ( x ) = f m - 1 ( x ) + β m G m ( x ) c 2019 The Trustees of the Stevens Institute of Technology

Subscribe to view the full document.

Bagging and Bumping Bagging and Random Forests Boosting Methods Boosting Trees Boosting Trees As a reminder, once we have partitioned a space into regions
• Fall '16
• alec schimdt

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern

Ask Expert Tutors You can ask You can ask ( soon) You can ask (will expire )
Answers in as fast as 15 minutes