You are on page 1of 8

AND KESEAKCH

Sector No. 29, PCNTDA(Ravet) Akurdi, Pune-411 044.

Academic Year: 20 -20 INTERNAL EXAM ANSWER SHEET Revision: 00

Term - /M
Department : AT DS Dated : 20/11/2019

Exam:
Ma PrelhmS Class : Div.

Subject 46.
Machinc lenin Roll No: Date : 30|10|23
Main Ans, Sheet: Supplements : Total :
Name and Signature of Supervisor :

Q. NO, 1 2 3 5 6 7 8 9 10 Total Out of Sign


MARKS

C.O. 1 2 3 4 5 6

MARKS

Start writing here and write on both sides

Hiexaxchicad chutering methad used a

data analysis to gra up similar data


points into dusts x
Of dustovp it reates
dustershierarch where
hi hex levels of the
heirarchy
Jevels
Contalns custers fyom owex

Agglorlomexative hierarehicad clusterin (AH)


Connnnonly used alqorithne that
stacts with each data point ay its
clusteys and then iteratively
the closest custessikxatively
Joaether merga
antil
single custere
AHC workinq
Tnitaliaion.
Merge cdasext clutens
Recursin
O Hiexaxchical shucture
Iypes O Aqgloxoreenatiye
Divisive

data points

a lusterin is an ündixected technique


used In
data mining
paifexnc inidentifying
fo
seue Yal hidden
without the data
oming lpp the techniquel
Need of clusteming
O Data explorati on
|Anamaly detechin
Recamnedation Systems
Imaa & daiaut TeLogniticn
Customes seamautation
Sevead types
k-meang
of dustexin y follows:

Hietarchi cal
DB SCAN
Agglomerative
Mean- shift
GMM
Fuzzy ducBexing
Sputal dustesinq
Q4a Random farest esernble learning
methed machinelearning it
is primayily used fox both lassificatioy
and eqression tayk
WokElow:
(OBaotshrapping' RE starts bs
by cxeating
muwtiple Tandom ubset oF training data
Decislon tree' foy each subset of datq
a decisio tree is Ccnstructea

o Randam featuYe selechen: When buiding?


each decision treeOnly a xandam
subset oF each geatuxe is cnsidexed
at each split
Vohing Quoxasin Once al the
indi vidu decisiomtee are but
are use to make predictioms thet
for
classification proble
Ady

9 High acuracy
Robustness
Feature importan ce
O Handle hiqh dimentional data
3) No
necd fox dahur scaling
b) Adaptive Baostin Adaboost).
Adaboost shaxt foy adaptiveboosting
ML ensemble tezhnique that is
used pximanily for classificatiom
problem

Weak
classifierl

teak
dassife 2 Votee

Jleak
dassifien

Nom- inegr
cdassifier
7 8 10 Total Out of Sign
1 2 3 4 5 6
.

RKS

1 2 3 4 5 6
.

RKS

Start writing here and write on both sides

I) XG Bost: Optimi zed and scalablemachine


qradieat
boosting librany desianed Fors
learnin task It iso widely used
for assificatien and reaessin problem
XG Boost

Levelwise txee grautt


Envronrment Qutput
Input Raw
Best
data Eesdavd
achicn

state
Agust
ment leaxni ng is a subfield of
Reignkorce om training
that focusel on
machine learning sequene of decislon
to make
ayut maxi m2e a
Cummuative TewaYo
to
environment
in

key elemeut

Enyionmet

State
Aion
Policy
umwative rewal
7 Q-value

Real time applications:


OGame playing
Robatics
(3 Atonomous vehicle

Reco mmen dation systm


Health care
fingna

D Supply chain manaqkmet


Enerqy managc meut
Petommendatien d pessomaizaton
b)
b Q learning is a model free, ofF -policy
Yeinforcemet learning algozith m uspd!
earn optima action slection
policies markov decision pTOcess
key concepts
State (s):

Action (A):
State achiom value Jundion (9unchon
Alprithm stegs
lo Tnitialize Q
a yalue
Select an aution
Tateract with the enuion ment
Update Gvalue
state transihion
Repeat
Policy extrachion
Exploratiom vs Exploitation The epsilon
gredy strategy
balan e exploratib
is Cammonly used to
and Cxpkitation
duving action selectiom

atole
Aetualud
Stote

Action

gyalue
Atr

stote >Qvatue
3gvatut
Ac N

Deep qleaining

You might also like