Week2Student2009

# Week2Student2009 - Week 2 Lecture 3 General m Model: Let Y...

This preview shows pages 1–3. Sign up to view the full content.

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.

Unformatted text preview: Week 2 Lecture 3 General m Model: Let Y 1 ; Y 2 , . .., Y n i.i.d. on [0 ; 1] with density f 2 F , F = n f; R & f ( m ) ( x ) ¡ 2 dx & M o Goal: Find b f such that sup f 2F E Z ¢ b f ( x ) ¡ f £ 2 & C M n & 2 m= (2 m +1) Find a Kernel K such that Z K ( y ) dy = 1 Z y k K ( y ) dy = ; k = 1 ;:::;m ¡ 1 Z j y j m j K ( y ) j dy & M 1 Z K 2 ( y ) dy & M 2 (Note that K may not be nonnegative). The bias part is E b f n ( x ) ¡ f ( x ) = Z K ( y ) [ f ( x + hy ) ¡ f ( x )] dy = Z K ( y ) " m & 1 X k =1 f ( k ) ( x ) ( hy ) k k ! + ( hy ) m Z 1 f ( m ) ( x ¡ shy ) (1 ¡ s ) m & 1 ( m ¡ 1)! ds # dy = Z Z 1 K ( y ) ( hy ) m f ( m ) ( x ¡ shy ) (1 ¡ s ) m & 1 dsdy The Cauchy-Schwartz inequality implies ¤Z Z 1 K ( y ) y m f ( m ) ( x ¡ shy ) (1 ¡ s ) m & 1 dsdy ¥ 2 & ¤Z Z 1 j K ( y ) j 1 = 2 j y j m= 2 (1 ¡ s ) ( m & 1) = 2 ¢ j K ( y ) j 1 = 2 j y j m= 2 f ( m ) ( x ¡ shy ) (1 ¡ s ) ( m & 1) = 2 dsdy ¥ 2 & Z Z 1 j K ( y ) jj y j m (1 ¡ s ) m & 1 dsdy Z Z 1 j K ( y ) jj y j m ¢ f ( m ) ( x ¡ shy ) £ 2 (1 ¡ s ) m & 1 dsdy . Then Z ¢ E b f n ( x ) ¡ f ( x ) £ 2 dx & h 2 m M " Z Z 1 j K ( y ) jj y j m (1 ¡ s ) m & 1 ( m ¡ 1)! dsdy # 2 & h 2 m MM 1 = ( m !) 2 . 1 Now we consider the variance var & b f n ( x ) ¡ = var 1 n n X i =1 1 h K ¢ Y i & x h £ ! = 1 n var ¢ 1 h K ¢ Y i & x h ££ ¡ 1 nh 2 E ¤ K ¢ Y i & x h £¥ 2 = 1 nh Z K 2 ( y ) f ( x + hy ) This implies Z var & b f n ( x ) ¡ ¡ M 2 1 nh This gives Z E & b f n ( x ) & f ( x ) ¡ 2 ¡ h 2 m MM 2 1 = ( m !) 2 + M 2 1 nh If you choose h such that h 2 m MM 2 1 = ( m !) 2 = M 2 1 nh we then have Z E & b f n ( x ) & f ( x ) ¡ 2 ¡ C M;K n & 2 m= (2 m +1) . Questions: ¢ The existence of such a K ? For instance, we could &nd K with support in [ & 1 ; 1] . The reason is as follows. Consider L 2 ([ & 1 ; 1]) , and a subspace S generated by ¦ y;y 2 ;:::;y...
View Full Document

## This note was uploaded on 11/06/2009 for the course STAT 680 at Yale.

### Page1 / 6

Week2Student2009 - Week 2 Lecture 3 General m Model: Let Y...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online