Machine Learning 3. Nearest Neighbor and Kernel Methods - ISMLL
Machine Learning 3. Nearest Neighbor and Kernel Methods - ISMLL
Machine Learning 3. Nearest Neighbor and Kernel Methods - ISMLL
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
<strong>Machine</strong> <strong>Learning</strong> / 2. k-<strong>Nearest</strong> <strong>Neighbor</strong> Method<br />
Error rate for nearest-neighbor rule / proof<br />
Now let y ∗ (x) := argmax y p(y|x) the Bayes classifier:<br />
∑<br />
p(y 0 = y|x 0 ) 2 =p(y 0 = y ∗ (x 0 )|x 0 ) 2 +<br />
y<br />
∑<br />
y≠y ∗ (x 0 )<br />
p(y 0 = y|x 0 ) 2<br />
≥(1 − p ∗ (error|x 0 )) 2 + 1<br />
k − 1 p∗ (error|x 0 ) 2<br />
=1 − 2p ∗ (error|x 0 ) + k<br />
k − 1 p∗ (error|x 0 ) 2<br />
because the sum is minimal if all p(y 0 = y|x 0 ) are equal, <strong>and</strong> thus<br />
p(y 0 = y|x 0 ) = 1<br />
k − 1 (1 − p(y 0 = y ∗ (x 0 )|x 0 )) = 1<br />
k − 1 p∗ (error|x 0 )<br />
Lars Schmidt-Thieme, Information Systems <strong>and</strong> <strong>Machine</strong> <strong>Learning</strong> Lab (<strong>ISMLL</strong>), Institute BW/WI & Institute for Computer Science, University of Hildesheim<br />
Course on <strong>Machine</strong> <strong>Learning</strong>, winter term 2007 23/48<br />
<strong>Machine</strong> <strong>Learning</strong> / 2. k-<strong>Nearest</strong> <strong>Neighbor</strong> Method<br />
Error rate for nearest-neighbor rule / proof<br />
Then we continue<br />
lim<br />
n→∞ p n(error|x 0 ) = 1− ∑ y<br />
p(y 0 = y|x 0 ) 2 ≤ 2p ∗ (error|x 0 )−<br />
k<br />
k − 1 p∗ (error|x 0 ) 2<br />
Now<br />
lim p n(error) = lim p n (error|x 0 )p(x 0 )dx 0<br />
n→∞<br />
∫<br />
≤ (2p ∗ (error|x 0 ) − k<br />
k − 1 p∗ (error|x 0 ) 2 )p(x 0 )dx 0<br />
=2p ∗ (error) − k ∫<br />
p ∗ (error|x 0 ) 2 p(x 0 )dx 0<br />
k − 1<br />
n→∞<br />
∫<br />
Lars Schmidt-Thieme, Information Systems <strong>and</strong> <strong>Machine</strong> <strong>Learning</strong> Lab (<strong>ISMLL</strong>), Institute BW/WI & Institute for Computer Science, University of Hildesheim<br />
Course on <strong>Machine</strong> <strong>Learning</strong>, winter term 2007 24/48