# jan26 - STA 414/2104 Administration Homework 1 on web page...

This preview shows pages 1–12. Sign up to view the full content.

STA 414/2104 Jan 26, 2010 Administration I Homework 1 on web page, due Feb 11 I NSERC summer undergraduate award applications due Feb 5 I Some helpful books 1 / 35

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
STA 414/2104 Jan 26, 2010 ... administration 2 / 35
STA 414/2104 Jan 26, 2010 ... administration I collection of tools for regression and classification I some old (least squares, discriminant analysis) I some new (lasso, support vector machines) I statistical justifications: loss, likelihood, mean squared error, classification error, posterior probabilities... I statistical thinking I framework for analysing new tools 3 / 35

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
STA 414/2104 Jan 26, 2010 Linear regression plus I variable selection: forward, backward, stepwise, all possible subsets I comparing models: adjusted R 2 , C p vs. p , AIC 1 and variants, K -fold cross-validation I shrinkage methods: ridge regression, lasso, Least Angle Regression I tuning parameter (amount of shrinkage): validation data, K -fold cross-validation I derived variables: principal components regression, partial least squares 1 C p and AIC can be shown to be estimates of expected prediction error 4 / 35
STA 414/2104 Jan 26, 2010 Predictions with smoothing regression I on training data: x ’s are centered and scaled when fitting Lasso, LAR, and ridge regression I ˆ β 0 is not included in shrinkage I (3.41) ˆ β ridge = argmin β X ( y i - β 0 - p X j = 1 x ij β j ) 2 + λ p X j = 1 β 2 j I (3.52) ˆ β lasso = argmin β X ( y i - β 0 - p X j = 1 x ij β j ) 2 + λ p X j = 1 | β j | I ˆ β 0 = ¯ y = ¯ y train I for predicting a new y 0 = ˆ β 0 + p j = 1 ( x 0 j - ¯ x j ) ˆ β j : use ˆ β 0 = ¯ y train , and ¯ x j means average for j th feature on the training data I see construction of tx and mm in Table33R.txt this is built into predict.lars 5 / 35

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
STA 414/2104 Jan 26, 2010 ...predictions > options(digits=4) > as.vector(predict.lars(pr.lars,newx=test[,1:8],type="fit",s=0.36, mode="fraction")\$fit) [1] 2.094 1.443 1.780 2.292 2.695 2.009 2.283 1.731 1.969 1.694 2.588 2.486 [13] 2.712 2.492 2.554 2.345 2.053 2.954 3.189 1.976 2.946 3.035 2.746 2.550 [25] 2.636 2.697 3.135 3.095 3.233 3.698 > as.vector(tx % * % coef(pr.lars,s=0.36,mode="fraction"))+mean(train\$lpsa) [1] 2.094 1.443 1.780 2.292 2.695 2.009 2.283 1.731 1.969 1.694 2.588 2.486 [13] 2.712 2.492 2.554 2.345 2.053 2.954 3.189 1.976 2.946 3.035 2.746 2.550 [25] 2.636 2.697 3.135 3.095 3.233 3.698 6 / 35
STA 414/2104 Jan 26, 2010 Derived features § 3.5 I replace x 1 , . . . x p with linear combinations of columns I principal components from SVD are natural candidates I X = UDV T , U T U = I N , V T V = I p I z m = Xv m , m = 1 , . . . , M < p I z m are orthogonal by construction I ˆ y pcr ( M ) = ¯ y 1 + M X m = 1 ˆ θ m z m I ˆ θ m = h z m , y i h z m , z m i = n i = 1 z mi y i n i = 1 z 2 mi I inputs should be scaled first (mean 0, variance 1) I Angle brackets notation explained at (3.25). Exercise: ¯ z m = 0?? 7 / 35

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
STA 414/2104 Jan 26, 2010 ... derived features I closely related method Partial least squares I also constructs derived variables I widely used in chemometrics, where often p > N I see § 3.6 for discussion 8 / 35

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
STA 414/2104 Jan 26, 2010 Linear methods for classification (Chapter 4) I inputs X = X 1 , . . . , X p (notation x used on p.101) I output Y takes values in one of K classes I output G is a group label: values 1 , . . . , K I response Y as needed ( Y G ), e.g. Y = 1 , 0 as G = blue , orange Fig 2.1; eq. (2.7) I data ( x i , g i ) , i = 1 , . . . N I goal to learn a model to predict the correct class for a future output, based on inputs code is in ElemStatLearn.pdf 10 / 35
STA 414/2104 Jan 26, 2010 Linear methods I rule: G = 2

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### 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