0% found this document useful (0 votes)
76 views18 pages

ML Unit 3 Notes

1) Bayes theorem provides a way to update beliefs based on new evidence in a probabilistic manner. It relates the conditional and marginal probabilities of events. 2) Maximum likelihood estimation finds the hypothesis that maximizes the likelihood function, or the probability of the data given the hypothesis. 3) For a machine learning problem with training data, the maximum likelihood hypothesis is the one that makes the training data most probable. This can be found by choosing parameters that optimize the likelihood function.

Uploaded by

Jaya durka
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
76 views18 pages

ML Unit 3 Notes

1) Bayes theorem provides a way to update beliefs based on new evidence in a probabilistic manner. It relates the conditional and marginal probabilities of events. 2) Maximum likelihood estimation finds the hypothesis that maximizes the likelihood function, or the probability of the data given the hypothesis. 3) For a machine learning problem with training data, the maximum likelihood hypothesis is the one that makes the training data most probable. This can be found by choosing parameters that optimize the likelihood function.

Uploaded by

Jaya durka
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 18

Uit-1C

Notation
1) Probabii ty -

Phobabit notahon s ofiuent wcu o witi


tue pobabtaty 6 evees hoppewine to1 not hapenif

TD we Qot noteuton s i t h Js wud for venn oliapram.

PCA) Bet A vent A happety


PCAD tomlemont- the oh.abiuty 8 ereu- A
hot hoPpeu
*PAUB) union - tRe prokibi lit e6 event A e

haprenif.
*PAnB) Treenehon- t e pvosabi uty
A tod B happeni

e. chonuws for aun e e n t


PCA) =
totol no e. Chon

Probabiiy Enanpte:-
Head Tail
PCA) n CA
ntS)
nA)=
S-,T A
Erent ef getltip heoo
nS) 2

P(A)-
Dia
1,2,3, 4, St} n (9) = b

A-Evet kor ebtivp an een humube


A-{2, b)
CA)e 3
PCA) 2 %
Conditional probahiiy
Two eency A> B
obabyof event ,a
avniup not
+ eet A \a
allleolby
appeol P (A) PAnB) A)1
PA
Eawpe:-
Dic

S={1,2,3,4, S,Ly
rosoh Lub or sotlup en
AM khen 4.
no.
shiun

Eren ho 4
A-Rs ttan 4 = {,,3}
=

B/A) Y
ThdepRncleut eneti:
A -
Idope nols ud ereuts
PCAnB) PCA) P(B)
=

bA) PC) p(B)


PA)
P(D/A)= P(B
2) TrueMuL-
Jnto a
docdapoivds
T R pottl oß vunnurp
uulale a outpu
model to Col
mawine Seakni
numeiCal Store w rotus o u o
mun a Mnle
Aeteted to ou mathieleatuip Mod e.
thou two types
Induutive
Dedwutive

3) Tdepevolewue
g t meQ ns ravddom number tod veuRi able
+hot s not afeted by o ttuet vaRiable.

*types Coolitional independene


* Untonliional Tnolependenu

ABoyejmle-
Boyes mule ovicles w with a
uay to upolate
OuuR beti efs baned Dn am va o he, rel evant pro s %
eviden Le.
Fomuula:- uketiuoDd
, PiY
PlA/B) PBA) PA)
PL8
Peterior Masgíal
)Pouter ioY pro bahiluy: vpdouked probahbiuy oter
the
ewiden toniduco
2) Phox probabi
lty: the bYobahiu y helr te
enidenu s (owidennd.
3) ikuu hood: PYo bahi i y o the evi
doute, 9qiven
ven tae
ta
tru
A) Maljinal probahiuty: Probabi üt o f a evideuu
nole awy CiruMtanu.
5 Bayes Theorem!

SMopAs
Dayes teorem
Fornula
Proe 4ee Enplanation
kEnarup&.
Payeteorem
*Payès 4ho.ven prop oea byThonmas Doyes
Jn IS Cenw ,

event boLoon
mor lcnbtoledre et conolihons.

Lketihood

PCA/B) =P(BA) PCA)oo


PCB)
hypottueM Masginal

Poe
et ns takL 2 eveus- AM
PA/B) -) Conolutio nal pvo babè tity
PCA/B)-PCAnB) PCB/A)- PCBn4)
P(B)
PCA)
PCA/B P C) A nB) O
PCB/A). P(A)= P(BnA) D
trom O
pCA/B). PCB)- P( B/A). PCA)
S e
PCA/B)= P(B/A).PlA)
P(
Teim eNHANotm
A potees Ae
B iveen data
se
PLA/B) firettnp probabellty hypo theu
Probaht i traininp eanples ove tven

FB/A) Pind prob. e. 9iven dotea povidtd


Prob. e upoteis eot true
PA)Pob 0. fotueis be)ore obeeviny iven dlea
qiven daka.
PCB) Prob e

Eamk- fa J, K,
PCkiny/tau) = Plfau einp) PCinp)

P(fae)
=Ix

12 -s
2ayeuian Leaauirp: Manimum tel hOO
keont Bquoreo exlor ypottQLLs
e (vobaoi uy
Boues ttheovem loluulok
e a t poti ble ypo theui H o e m o pYrokade

Maurmum Jiluuihood heart pualod egOV

wyrotwis:-
TO fis ta maumum iketi vood ypotheis
bay eioun Aol nAnp
lp)a
(o) h -
Y max p(h/D)
Map
heH

CANT
Aot s Adke rouuin JnstomM ( , M2 Kn) k
Cowidu toasget vouluus D- (dn,
d1 owm)
w e wmle
PDhu) os proclu e
p.(oli/w
wap = NGman
p(oli/u)
ey ovwmwp noYmal odishnbuti o
e (-H)2 -pro olut
20-2
map Cuma
e 22AI-)2
heH
2T
Orma VM

h e
Edi-h(x})
we o Junuti on
a s mon Jog 2g2 (di-n (x;))
heH V2ta

di- h())
2
M
a r Mn oli -n (xf))
heH

=rain dt- h (ni))

m
hmap bY ML aYSin d i -di- h(ni))
heU

T t a eay- Auord eLS hYpo t

ELaple t-

The tet yokwms a Coreut pecitie cult un ml 28


Hhe Canes in voith t dllear s atnally
Procen > anel a loreut nopatine e R t D my A
O +he Cones u h 4o oie a t is not preeut?

u t u huoe, O08 e Oute opulobn hoe Cante.

uyesaoem PCh/D)= P(D/u). P(h)|


PlD)
PCCameet) o0& tCoLe)
=

*9712
PC+lcanu)- . 98 PC coun uel) =02
P+/notCanet) -.O3 PC-t noECon w)=
9
Plecnes/+)= P+/Can et) P
Canut
PCD)
=O8 xo.ooe O018
PCb) OoO8
PCD)
= O 2

PlComu /+)- P+| O.299)


D Conuu)
PtD)
pCnetCanuu
=O x 049 2
6.290
PCD) Po
292

(o O698+

AmL w not amuer


Maximum iketihood Hypo-teeur or ect
Paobabilhes
Conider tLOth use w to
tee ettiue
e c n a honeletevnuinutc ahebi Listýc)unth,

+ o.1y, stun has too d& ate P va0


skece
Ke oat a fun tíon approxinnator
Output is 4 proeabi ty d o ) =)
o f t wordy lecctn tte taRpot funth
[ o ) Suh that f (a) = Pf(t) =1)
How Com we luaun p u a weuaal niitwo le ?

brde fome nay wold be t tw


eLcetmoe epunies o 1's o's for
CoRCaut Fo
Caun pemble ral 8 o ttuon, sa t
housalwtwoe bo eutpuut thL teipe enu
4ideaion molel Ne opti ze n 9Rdel to s nd

kai hood r o A TFin thb M

i t otaun an enpreMm ° r P9(n)

eom
tdi)
**
Xm dm) Lshe oli i t
lhtenne O Coy Va orf (i)

o t X di anolom rasiabes,k
rouininp enampe s doc
OMAWMup hou ealh
os
Can omte PlD/u)
n9e
Sdopondekuj,
PCD/h)= P 7}, oli/k) 0
r produut w
Apuuy
b/h) - PlAi/A X,) p lxi)

&e psab ity P (Ai/h, ni ))

pCd) h oi) hCa if di-


(t- (ai )) fe di=o

Re e p e s At n a more Mateonocti cally


aupuloble o m as

PColi (h i)=h ( ( h nj) ) O


ERuaton G) toAuubhte Jor P Cdil u aj ) in
epuat om ( ) to otato

PCD/) =

h)dt (t-u)) -lo:


P(Mi)) ->
we wmte an enp-euion lor t Manim
mun
ilelihood wyro thes

huL =
a max
AhC)dt (t-u i)plx
heH

8
Boos epti mal Loui hes?
Snopdu :
ef
*Baes opthmal ouci be,

Fomula

Enompl
ouyes Optimal amfie
as
TR Baye opbmal ttoihe
obchiAh mpellhad nAakkoa mect pYobakle
reicho or a noo eople.
manirom
A petett or wmou

Aprdeobiute aneoore epeuolto u


MAP 460
rd +C neprobole upotUs
rora raininp dataeet

T
wot probable dami hcahin eftta
NOL Wtan obtuneel b Conbiwin t
reli cho ef oLl hpo thenls
hoir hMertor obalitities.
weifhdeol

Fo ua

e e oph mal Aoni heatton

Ptvs/D) =
2 P(y/) /D)
arrnoa
P(Vj/hi)plhi/D
Vev hiet
T e epn
eplaias tous o toluulate
Cevoletional vokahili ty or a nowiwteunu lvý)
kom Aome At v. ven t e
Toaiiw bato (D)
en

Te optimal l i e o hws

PC/D) manim vm.


Exauple
e Hq pottekpaue 4= {hU h2 h3l
tenioY PhtD)
Pexteio =
o.y P (ho/n)=o.3
plha/D)= 0-3
hO4 h2O 3 hse O3

wa wpottes
SLppou a hO AVMtanu
T
T h e probaity hat x pou tive is oy
Th
h h
prooabiit that it u nopotine is o
D3 to3)h2= O 3) h3 o 3) Sne=

Henu, nuot poaLle Loi hLa) on


As napotive.

T e Let of peu ble louni h'cahon o6 t

veo
PhrlD) =
O.u, P(efhr)=o,P 0/t) =1
pCh/D) O:3, Ple/hm) =1,PC elha)= o
PCh3/b) - 0:3, P(o/h3)=1 p(0/h3)so
halore
(e/hi) p (hi/b) P(® (hu) P(ht fb) +Ple/h2 ) plh2/b)
=

nie
+Pleh3) p(hB/0)
1* o.4)+ (of o.3)
A

+ (oo. 2)

ha
hjEH Po/ui) p(ki/b) = Ple/A) PCA1/n) tple/k) Ph-/
+P(e/h3) pl h2/ID?
9 o o 4)+ (1*o.3) + Ao.3)
m . a L MUnM 6

Heuu PCyh)PCAi/D) = wepative


armox
hiEH
Naive. Poyes cahes ?

Naive Layes clouwihe


Fonmuula
Typ e Naine ay louhe
Advautepas
initaios
amp
Ocue ayej ciamife
O a i e ayes axuwmes Conoli tional
Undepenolenn bvor t r a i p dotauct
o claafe epes otes oloda int dijeut

oMs aubrdunp o A ay e tOoem

Puct cRumes +0at elotionthip b/


auu inpu eocw o a clod volepenoleM
Hewue, +hs MDolel Calle.ol OOUve (

Forwula:-
t ormila oile anump nolepend e n

es P(A|, .dn)= "d Mn|4) bCA)


theoe m
P(1.. Mn)
Oouve
P A J. Mn )= P(Mi A) P(12/A). pOAiJAJPCA)
TYpes es Natve ayes tauulAe:
P
einoulli Naie Pyes
ultnovial Naive Bayey
GaKcun Ocuive Baues
oauve ayes: -

i) Retwalli
eoictord OL ooean rogialle
* Topite for ois(gete -teatuy
Tf Mwino aU euve Payes :-

* dvepuany o prOuunt words 0 featws

*CDmouly we tor docunme lauel fation


Poolaww
tuy Gauuplan auve Paues
DAed wan dcle s as re de auuia
dutobuti Dn w
reolictos c Lntinow raicoB e3
Adveages
kEay t wole i t olon wup binaty Loy
Cokepor cal input Valuas.
Hnd les bo Cowtinow aliscuue dotn
F a t k Yelicble -fsr hmatiup realtime
Preditions.
Liwatabows-
*Ammd - a t all e feaknus oe Inolepenolent
9 i i ihly n ilceley in prosbca Scouaelo.

Uwuitabe for Leeical clata

Conputeltionaly enpentie nwhem wmed


clonm a tae o9e Tfeml
awple : play tenis
PlayTewnds taininy tnanupls
Outlaok Tewpeintuu Hw odi ty wwd play leid

Sunw Hot weak NO


2
ttoE No
3 Strou
Cellart Hot weal Yes
A Rain
Wea Yes
Ran Cco Nomnal Wealc Yes
Roun Coo Nonal
Ovelcast
o No
Cool
Nowa
Sunny ild
Strong yes
Jealc
SUww tocul Novmal
No
to wealc
RoLin Mid Yes
11 Nono
weak
3onn Mitd Novnal
yes
2 Ovettat
Yes
Ovecaut tHot Strov YeA
Norel
Ratn isit yes
Strons No

LectniY phall -

Outlook Play-es ploy- N Teperahwe Play= Yes


Sunny play- No
Ho
Onencon 4/5 O/s ild
2/s
4/ 2/s
Rain s Cot
Phy Yes lau- No
tumidi y Pla ye
Play No Wwd

2/9 4/s Stronf 3/4 s

Notmal / weale 6/4


Ptplo yes) - / y Ppoy =No) S71y

Text fanale
Giena new Initcme, prediE 74 Aabel

X= Cout Jeok Sonny, tenbeRatRe = 00, ttumidu


HiA whol Strov)
eamiup plhouA
Loot up tables aunieved tn t e
P(outlook Suny (play-Ye, l =%,P(outLoole =Sunng
play No) = 3/s

PCTepexatwul =cool/ pto= Ye3)- 3/9,P CTempetatuu =Coo/


Play No) = Vs

PCHumiwi) Hiph IpAo- Yes) 3/4,


=

P(HuiutyHh
playNO ) Y/s
PlwndStronplpo Yes)=7,PCwlnd Strnpl ptauye NO)=3/s
PCpto yes )= 1 /1 tgP(ptay- No)e S7
Deisi on aleuip wth t Mapwle
PHi9ye)
Plye/x)-P(Sunnyl yes) P(toel /ye)n P(Stron yea))
P pMoy= Y2)
J8S o oos3
PNO R P(unny| No) p(toel Ivo) B4i/%lro) p(JtroY]
PCplay N9
89 0.0206
Civon te Jart
x to be
Ye/x")<p(No/x'),
No
we abel
9 Bareiam Beiep newolc BRN)
Stnopis
*Bayexi con Retie ntodt&BA)
TwO importat onu-pt mup BEN
oeuteel ouytic Graph LPAG)
y Covditiol prdsasiihy taHe t
epD
*Eaple
eno Belieg Netwek (BEN):-
A L &
pobabiusti Groptcol Hools
a

hatepretes tonAthbnol dopenolenuues bo roundom


votiaHes -u A
obCfeol Ay ie Craph (OAC)
e4t aluo Autaba d
elatio w tiRa emeus
opreeutup pvokshir
a bru
&t 9OYla
on
supetvtsd oinp
o
depordlemu Srolhend n
9o Juportant (owLepts wup
seN:
eDveteol ay Ui Grreuph
s tDAU)
Covlitibnal Drcoahiut Tobl
Direte auy u Gapt CCPD e

(DAG)
eA cdA bwole s AQURal netwoxl
deep oouwinp t o
Xayes Oraned aa
A
A DA hetwolu am
La a
OAdhleetwtt owin Aauyed hae mort Cov
nplur
uthoa Joues bwtpuuts to mutipe loyesm
weus
Ratn
DoRLok/Roio
Dos baa
tottiol/Doskank
Cectnides
(pwdibiowal pvobahiuty Table CPD:
J t ds -fe probasiu ty ef Dne event 9ivon
ouRlenle Oo-ter enent, 0ten d exui led
+
m tetwk of events A Dn two dependent
adomn valiables,

(B-TR=T) /48= O.!9


R R
(BTR=¬)=) 14g = o.39
18/4
(B F 24s o.ob
R-T) =

B 8 1 (B-F R=F) L9/4 D.37

nampl

ugta Eathquak

Aou

,cauls Cas
o

Rien pobabi u fie o

P( T) = 0.oo

PCL =O199
PCE-T = 0.o02

P e F )e 0.79e
Propsaiu ty eß tlatm
Buuglas (B) Ecuta uake (E) P(A=T) PCA=F)
T O.75 O.o

T te O O)
O.99

T O.2 .T
A
o.99
Probablthi of P
Atoum tA) PCPhET)
PCF)
O.So O.lo
O.Os
.&

Pud t pvopabi lt Te

A T
ot P s T P2 A T
B F E
tie) P sFP-e
(PP2A~ B, E)
CPA) P(P/A) P(A
O.0 /B, E). (~B). Pt
B
O.00062

enolonl in Phtoutnotes

You might also like