fo da linear regressionjeffp/teaching/foda/lnotes/foda-l11.pdf · sam f squared errors...
TRANSCRIPT
Fo DA Linear Regression-( 11 •explanatory & dependent
variables
Data labeled
datsX.
HAITI! rations
XEIR" xd
GEIR"
(X,g) = { ix. is , I
,Huh )
, . . .
,
( x. yn ) )
X - explanatory variable
y -
- dependent variable
slope,
intercept
fit.ie?n--axtb!÷h
• Measure Error based on
Prediction( want to use on data
ftp.nqorasdependentwe don't have yet )
o ⑨-
←& fit l←
fiecxtg.x.is!
How to Measure Error ?-
residual I = llx ) hats !
ithp!:{ ri = yi - gi - g : - llxi )targe
,predicted My
Som of Squared Errors' ab " model
- Mal )
sseax.net#isise*siii?:?T÷÷::÷÷÷÷ri =3 - 4=-1ni
,
Sam f Squared Errors
##datpn
yo
SSECK.de/--&lriT--EICy..-ee..s5data model E- I plaguedprediction
whg?lab#Ty
• Squaring → makes non - negativewhy not cbs -value Iri )
• Norm ( he - norm) llrll !r= Chore.
. . . rn )
commonnotation
value Wi
predicted value Ii
a Start w/ Bayesian InferenceAssume Normal Noise on gi
NUE ),
re )↳ Negative Log - Likelihood
d*↳
sssccx.sl.hn• Easy to Solve
¥.sn?ti:as.:insseeas¥• →
" closed form "
→ gradient descent
pen↳ convex
'
explanatory ra!
t¥Is
a'
.
bit-
- aa? SSEKxstea.is)la
, bC x ) = a xtb
nI
. average I -= In II,
Xi 5 - I ÷,y i
~ Ra . Set "
Center "= ( q. - I
,a . I
,. . .
,xn - I )
-
- (,
- 54 He - 5 ,. . , yn
- 5)
. ¥i*÷¥¥¥¥÷b = g- - a I