O-Vams

Macbine learning 2eNUIA42

CSM
Assigrent
neighbor clasification wi
9Explain ncovedt
on erample
) Neanest Neighkour Cles ifcabin clars uting
Simpy by. leavnirg exomplke fer each dz)
decisien ule (n-n
the neores nejzhéou
t data set od onod is diick
Astuoe that we hawe a
into 2 Poxty bainig set
testing Set
> Ne have to design a clstife 6y iy the
traáning set data the aCtuea(y
Ato clastife, we wil chet k
class ifres by win test ola ba
of tle Scena iO.
This iS genehal neaest neklo
’Ale ane implenenbing the deign
classifte.
Propestie Hain prope, tiey of neoret negh bsuy
classifies
OFor exrnple we have training set Consisty of
identi cal inttan Cee,the tuo nstantee ae alnas t
Sane bot they Core fron diffeat classe
diffuat classey
if tte sane hstance belons to chsu anol tare
instonce Beleng4 fo chss2 - i/ this situntion
in this ase
15is is vey casy be Cawo te classific aldnd.
nemorier te toining data :
 w

a right exemp la
Gheesally cheyins to enenly
if is alnost neen to decibjon bounday in sch
Such that
Can nore or les! te prcsmt any this we
Neueet leighkour Cast iflo decision bourdoy.
- folley-Low bie', lyh
Ovenfitting
is limited
- psoblen occust thon -if tle t raining nore
meod S mal Set of data is dalea
(3 J/ ue cwent to Girer
tepresent te
sane neoreytcases
in tenet of Alguih mic point of vicy:
tie t of erer
Proben, i sboring the eremplou -bub thee is
Suprose we have a
"ine b hene we Con ireseose igle iny oe -sa)
classifiCo. tion e reduce te tine pemance
Le ofot
usig elobrated odeta dtrebue
i.e pefectly organi2ed tte erempla autoetiall,
tLe tine is eicel.
0e sinple problen i-if we have mrel hish
di ronsiona lity it il hare r ypawe
fig0: A|l tle § àhs ton Cee ane nener fo cachoe,

|tig: ke no of dimendiony inchsasa -t most of

the Space is Cmply, ony Jone pasitiong ore
fleck - this is Called spare e ae
Calcutatig pair ws disfances £ls
) Explain Hienachi cal clustenins
obh aneotire
Predictive cl
0nstance Spaceusteing
clustuing-i poxttn the
’ his Hienochi Cal chus teing is aolescrip tire
is
Clusteing Des Crip tire clus texing meany par tin
tle daba sathe tho jns bance
Spoa ce
dn Rc Jererally the clustey represen bel
in the orm b/ tree and these treey re
Galled olendogrnd
’Dendogra t
distance a
ane
pureb
meaSUe
defined rin tom ofa
>Dendogram is bree , especially it is a binay
baee tow te tree is ferned in tihe
Ifera tion the eleraty ef data Jebfrt
Greneral meadrment we have ce- distunee
Simply find tho
clusteI Clus ta clistance blu 2
clus buy mean
based on tis Valwe we
will hoc close
tLe tuo cus tey ae
no Prodvce
goo Solotion . peugtance doe
bot this

>Lintage feaction: L 2**2? IR

5is function, Cale ola tey tho dis tance 6l
arbib vay sibse ty f iab bance Jpace ky usins
tance metric
a a drs

| Sngle (inkage Fonetion:

Eaeh clusten Conbainy diffnnt set of poinby
Each and eveny Cade we are Condiderin,
arbih wng sesbset of, ins tonçe on Jingle Linkoge
forction it defirney kte
tle dis tance Slw two cluste
 mean,o
mern
DrsaJAl Lcenla,s):
ene bldistance
w 2mean {le
ofmeon cluste Cluste
stea teclut *-Mesn
of
Assue cittance teCaledate
ti inkage neter Thiclfonctn:
sasbetuees Lintoge (entrtod
Lavs
avergse. caleulate
te tlen Gnel
te, is pai evey cach l wdisfance tle
o hare
to-ee ce
tan pairwice
dis tverage
chsber
ds fte linkage:This
defney Avenge
Dis or(a,8) Complete
yCx
cluiter ech fon elemenby
trnce ds
stblw ance paiwse deest elusteny
at tuo
tle definey Ihs Linkagete byleto
b4,8) iggle Repvesenttion:
He frn distance Srmallest
itpoinb4 tuo the ting nec Con By
bly
distance palrwise e cluste cach
Alemr Clostening snalleat tfonhe o
p
 HÇerarc! i Gal
Alaorithni
|rput
HA cl
Costreni
Co,L)ng Alsorithn
dataGx ) linkage furction )x2
Otputia denologran sepresen tig a descriptire korniy
I oitiae csten to sigle to data poat
2. Create a feaf at level o fo eveg tingle toa eloste
3
epeat
find te parr cster ,y si lh loust lnkese
Creabe a parent of y at level d :
B: ontl al duta poinby ane in one Chsbe:
9- setn the Cont rutcd bigh tree with inka
(evetr
Advantaget
Comared 6o . mean
clsterig
2 Cost of Conpetition is redoce d
3. weneel not chood e te dibunce ,we Can
chose tte onkage tunetion alsO
Describe tte procedue sed 6y least Square
me thoet tr preoli etig target clase!.
A) keast auave Methodi
LSM iSis used to learn (ne odel for sega resio.o
and clasrifi cation.
|O Regses ion probemi firhng a function estinatr
fso m 'eranples Cai, fci))
fron Cai fa)
Resia: Diffonce betueon
etbd 5fuel esbina teel
functionalug iS alled desicdea
desidual actal esimatd
Leust Saae Method definition: 1e Jum of ypare
f sedtdua ic fo be. minimi ed llel LSM
4f te diffesce is vey lo w Bhon we accvrately
estimate tto vabue
LSM
’Unia iabe Linet eaesion
Constder twofeat re wCeight 2 Cheight)
is alrendy knos tle step in tecept cqwatin
J at br . e
proidig 'n'valu f l;,hi)
Hee tre assuniy a Univoniabe (inec enatir.
Similas to store inber cept eauution.
(wi : Cat bhi)
riginal Ytimated
fit porning
Simply the angwer S

Ja u-at64; )) D
valve abuktiy
Asture teate'e' foaet veniabele iy tn the
finq te (oriqne o ely
n ttee
n fle vnte of

intep t Teqrewin Coefficiet fer

inteecent a -te regreog ine.. 6le
2nd ptep pooces ith
we replbce
eplace wjbh
This is oo ma liwed feature
2) Take -Covariance etween tte nornaiel
feature bage t vasiable

Sum of besidunl
Cy; - (a tb )

Molkivaiate Linech Rogreion

Suppose ·if se have en as bibay nunbes offeatre
te have to we matr nofatioa
Ho Univari atbe pxoblen repeontation s vetel
inbo ma bri forn

bis a salan
Sinpify tledolution Ase ce using
Coordiate the entire difermt éaatlon cuehonggendoy
depresnbed
tingle forn ate

Rerainirs all vout - degre ssion Cocfkicie» b

Deplaikd Pepcesir
Ir
Corstrint to veigt t veclor 2 clse prfor

ptimisat io pollen

The closed frn solstien :

Uirg test Jagre regrecsion for classifitio..

ie eje ires res reisi to leern a kingy
Classifiteti
Poseramplee and Neg eanple,
t

Te leust ssuare Chesifie lecry te ecitker

bosdey
Cans
Cors truttior tf linea claife wifh decisie.
bourdoy Condhuctig at M
-9A)with
Cno f )with ditt possibe
posilo doice
The full Covariono pprta ch uilh tM_x'
bus tie onpleri by Olno).