clustering notes btech cse handwritten.pdf

Notes

1. What 7
Juste fuster analysis is
nsupervised learning technigue
4 の pre 0 5 x way thor
byes の the same oz の を (cluster
AL mo mila each other
than Lo tho ın の の 2 る
oaithn

-mean chu

ieranchical

Eoupectalion. Moximizalion. algw

> Density based clustercng

2. ®

K- means clustering

Algonithw
Randomly seleck k cluster cenbe
VI Ve
Step 2: Colewlate ho distance belween
each data point aj ond each cluster
Step a

Assign, 008 point a; to

the cluster, ce vi dei which

the distance lla Vell is minimum

Step 4: Recaleulate each cluiter contr by

taking the average of clusters data pais
Step 5: Repeat

P ん し step 2 to steps
until Ahe accalwloted cluster centers
の Az some as previous の ん No

reassignment of date point happend:

3. いい を 호
性 5
Distance between data poin
We as that each. data paint à

dimensional vector

he distance between two data points

is defined os

A TR 4 Om”

4. e

the cluster

Assigned
tot

6. ©

8. > each

observed

there

contatnina ‘all

y whether

decision regard

ano to be merged on not i
y ve

token based on the measure dissimi laute,

between. the cluster

10. 4 dissimilarity

do

Average
Linkage Tinkage Linkage

Pet?)

ee ea

Complete

4(182=mamfd(m9) : eh, geal
t dl

11. whslitchirn of

12. gglomenative Clustexing

E

distance matrix, construct a

hiernachical

, method.

ndegram. by

Step 1: a euch 2

ten, so that we have (N=5)

clusters, each containing just one. item

Data set = $ a,b, cd, e%

ret cluster set c,: Sah $05, {cf fab fet

14. Step4: Repeat stepz ond 3 untill all items
ane clustered into à single cluster
에 sige N

17. Divisive Analysis

ance of x

Move + objec. with the maximun

4 each.

D verage | gel
y = aegee】d(cg):geCit -

&) Compute
average$ (xy): ye)

り Find on objet x in G dor which De

If Do then move x to G

Step 5: Repeat step4 until al differences
Dy, ane negative. Then. Cy is split
into C and 9

18. 8)

19. 3

xpectaluon. - Maximisalien Agoaithw
cor -maximisahion algorith

© dind が

teal model

sta Ls)

egualions cannot be

variables and

hence

be solved olisecth using MLE

method

algoxithm

ん ose the parametas の to be

estimated.

step 2: Enpectalion step (E-Step)- Using the
observed available data of the
dataset, estimate guess) the values
of the missing data.

Step 3: Maximigälin step (M-step) - Complete deta]
genenaded adten the enpectadion step
ts used 加 update the parameters, 0,
by manimizing Likelhood fanction

Steps: Repent slep 2 and 3 until converge.

20. 9 ds
unlabeled dla pot

pao blero

algoasthm

Sor.

Gaus:

sian Mixture

to blem

Suppose we ant given a
observations Lt, Rz,
ne variable X

Let X be a mix of k normal

distoibubions ond Lek the probability density

21. Likelihood Sunctien

Mi soe, M (for

22. the means Ms, the

and the mining

HT

Recolculate the promos using

step 4:Evaluate the Log-likelihood Sunetiom
and check Sor convergence of either
the parameters or the Jey. Sikelihood
function Tf coveye ther stop;
Ehe ga step 2.