This preview shows pages 1–6. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: STA 414/2104 S: February 23 2010 Administration I HW 2 posted on web page, due March 4 by 1 pm I Midterm on March 16; practice questions coming I Lecture/questions on Thursday this week I Regression: variable selection, regression splines, smoothing splines, wavelet smoothing I Classification: discriminant analysis, logistic regression I Kernel Smoothing Methods; Model Assessment and Selection I Projection Pursuit Regression and Neural Networks, Ch. 11 I Support Vector Machines, Ch. 12 I Classification and Regression Trees, Ch. 9.2 I Unsupervised Learning, Ch. 14 1 / 16 STA 414/2104 S: February 23 2010 Wavelet examples Buckheit et al., About Wavelab 2005 From http://wwwstat.stanford.edu/ wavelab/Wavelab_ 850/Documentation.html Compare Figure 5.17 2 / 16 STA 414/2104 S: February 23 2010 ... wavelets Vidakovic and Muller, Wavelets for kids (Part I) 1994. From http://www.amara.com/current/wavelet.html : Amaras wavelet page > library(wavethresh); data(lennon) 3 / 16 STA 414/2104 S: February 23 2010 Ch. 6: Kernel smoothing methods smoothing without basis functions I model: E ( Y  x ) = f ( x ) (smooth) I data: y i = f ( x i ) + i I simplest possible estimate of f ( x ) = E ( Y  x ) : I f ( x ) = ave ( y i  x i N k ( x )) running means I N k ( x ) set of k smallest values of  x i x  nearest neighbours I weight cases according to distance from x f ( x ) = N i = 1 K ( x , x i ) y i N i = 1 K ( x , x i ) ( 6 . 2 ) Figure 6.1 I kernel function K ( x , x ) = D  x x  or D  x x  h ( x ) 4 / 16 STA 414/2104 S: February 23 2010 ... kernel smoothing I determines the width of the neighbourhood, hence smoothness I increasing gives smoother function (higher bias, lower variance) I constant (metric) window width constant bias, variance 1 / local density I nearest neighbour window width h ( x ) constant variance, bias 1 / local density I Choice of kernel:...
View
Full
Document
 Spring '09

Click to edit the document details