AssiGNMENT -DQ.

1.
explain dlferent tyfes o! Boostrg a lqothm?
A-Boosting
Bostng is an en semble leanng technigue
ensemble
Hhat Combines outtiple weak learners toCreate a
learner, Dt woy by sequeraly
model focuses
model1. where eoch ne orrechng
the eors ofJhe previou modelt.
-

* Totaluced by treund and scha pire.

weigtts to m's clauiied.Sample
* Asagn higher
bains
weak teorhers seguerHaly
Jand
uses
Decisoo Stumps (Dne - level decisi'oo to
tee

as weak learmes-)
Mathemo tiol coneept '-
eachweak learner s auigned eiz ht a baed
on its accutacy
et

prosir
Sample and implemert
 Cons sensihve to utiers
Lhighweiqht mutelaufied po»ntJ.

> Gradlent BoDsHra

* Thstead d weighkng mis clauitied points, GBm
mioimizes the resideued e rors
descent
Uses mean Squared eror, Cmse) oY
functo.
la the los
Pros:
obu st than Ada BoD

orcs ucllom com plex , data set.

lons:
compu tatHorally expensive
Con over ft without regu lorizaton.
xGB00st lextrene Gradient

|# An optirnized vertfon ot Grad'eot B0oshing eith

Her. performance and speedt
be than GBM.
parallel roceung make f faster
Valuer , bbiienty.
thnds

Prosi
tastand. s Calable

y avoids overfimig uig egulavizatin techniques.

|4. tght GBM (lght Gvodient Boostig)
* Developed by microsoft, a ster thon XG B0ost Por
|lorge data set Jostead level-wse.
uses leot wtse splrng
Pros?
faster han x BOOst
uses les mermoy
Cons
not wokwell wth Smol datosets.
may
more Sensive, to over khg.
S. CatBoos ccolgortal Beodig
Vander ,
Developed byVandex Lpeciicoty designed to
|handle Cateqortal eatures ebtitientty.
to avord overtthhg on Sonal!
t uses ordered. Boostng
atases.
ros:.
data,
Best -fos hordling Calegur'cal
woks el! with both Snall Ond lage dateeh
cons
slouer than ght &BM on nunmenal deta
.
Compare and contsast Bgg trq and
adventages and dis ad van tager.
featureBaggng
Tains multiple
Boosting
weak modey Txas wcak models
koncept inadeperdenty inio farallel 4 Sewen trall a where eæh
eaeh mode
aveages their regu lt. Corrects error öh the previauy

Goat Keduc es Vanance to inprove Reduce b'as to improle

stabilty and prvent
OVer

Tainig each s toained indeenc Cach model is toained to

|shelegyy on ditterent randon focus on misclay ihed
Subset. Pants fom the Previay
|THeration.
mode)
Conbinut
uics majoriy Votng or uies a weighted turn ot
model1 qking more
averagingiov o -to hfqh pertorming ipotore
rmodet.

|Advantages and di's ad vantag es

Bagpg adiant aqes?
Red uce s Vanance and movemts Over H*rg.
high - anance models thbe
e
well wth

decsin tsee.
Suprort povallel proceuig.
 Ds advamtagei!
Does not reduce bas Coot Suitable fos underfihihy
models)
Not ideal for corn plex poterng requihg seyuential
Cormeetim.

Beothng advantageei
leading to higher aenay.
Recutes wel!b'as,wfth uweak leorners by too pro ving them
wonti
|Seguentatly
mpre elfective for complex datarets.

Ds ad vantages !
Prone to oVerthq
rEgulonzed Prypesly.
|computatbnay expensve
expensie dee to
due do seuentialy tooihing
ore sen sitive to no'sy data,
3. ¬rpli 'a ctive leaminq with its heur'sti and
Quey stategie.
Active leanig: - St isa m approa th is a model
selectvey quenes the most io-foroathve dlato otnt
|Por rother thon reyng on a -aly bbeled
dota set.
This is usefeul wshen labeling data is epen sive
b time- consum1hg, Sueh ay fn medical d agno
Conneton.
 Tosteod on | vorbble dota, achhe
flentiies the mogt uncertain or sefel nsae
leamig Oracle to label th em, leodng to
and
beker perfomnaoce eith fese labelled somple

Heuriste in acive leoroig

Actve learning relies n diterent heunst to
deide which data pofots o ueny level. lett.
I. un cerainity saro phhg :- selects Samples where
he modle! s mos t cuncevtai.

|9. iver sity - Baied camplhg!- selects sam ples that are
diverse kom qlheady bbeled cple
3. Expected mode! change: choase onkthat mot
signifotty inpact he model.
4. Suery -by commi Hee!-yes mu tsple modelt 1
to
Find poiots here they dis agee the mot
Query stralegies in Acue learm hg!
Ahve leanihg eloyes dihernt uey
Stootgies to detenihe hizh ata pont to label
next the types ot steatgiey.
 cevta"y
* The mode! select ostances wh he highest
tuncevtainty in PredieLbn.
* least- Cooident ;- pick the sample where the
|oodels.hfahesh predicted probability is
louet

- avg min P(f)

avy mi% ( Prylz) -p(yl))

Quey by -commiteel
ned on the same
model stoaained
multipe
uses
obelad data but asith.
with diterent ayumphny.
selects samples where he cornmitee dsagree
the nort

* D}agreemenk maUrey.

Vote enopy

3. Eypecked modelcharge
seles sameles tha la beled uwould

the model's Parameters the nost

 coroputcs the qadient the Ibey fuochion 4
Samples sith laget eKpect qradiend. choge
4. uersity- Bared Samplrg:-
eosures hat he selec ked Samples
the are

Aiverce and not sioily'to ach other

Uses. clu steng technigue.
k-means, Density -Bued Sampl2
ex!- labeled Poiot to
prcks samples tor exishg
eatuc space.
auto encoder? explaio Au toencodes
what s
neat diagrami?
with atefal oeual
a'type of.
An Autoencodes s leany
unsupe'sed lomhg
beto cued By
ndthen recon stuct ioput data,
a
to com preu .bluw the orghel and
-fhe diteence
encoders are c¡mm only
reconstot ched data. Ato
ued or dimensioralihy reducton , featere
aly
extachom, anomaly
delec hin, and,
or
derothg
data.
Styctue of an Autoencodey
I·eneoder!
compreu es -the nput into a
lower ditnensInl
eprejentation
ues connected
layers Con volutio layer
depercng the dato type.
| . latent SPa ce (Bo e neck) "

A- comopreled repre lenatim the put

+ Captures the most im potant Peotres ohie
temoung noße.
3. Decoder
. Re constucts the orghal input horn
the
Cormpreted repretentation,
|. shold
Shautd Drocuce output as close a Pos'be
to the
Drginel nput
"ypes ot Auto encoders
. vanfla Auto ecDders ?- w
Barc Stau cture
a singe b'dden layer.
a- Denoish Auto encoder
Toained to remove
norse hrom iput
data.
3. sparse Autoenco der ues sparsity Constaktf

to improve teature teamng.

y. vanotfonal Autoenodei! Genera tes new deda
shnilas to the npak eued io gencrali-e rodeling)
s. convolutionul Acelo encodeys- ues cNN Laye toy
baled eature erteachn.
latent
spale BoTeNT

AuDe DB077LE
Chect

loten t
Space
An(ODER

loten! peectei
 ¬tLAT algo tm.on the -follawng deBaset to
R the uent fa Hery cwith mt suppot

Bread, ButHer, Jam

Te00 Bute , cote
300 Bu tter, milkk,
Tuvo
Breod,Butk, coke
Brcad, milk
TSDU
Brccd, mi|t
Sread milk
T400
t800 |Bcad,BuHer,milkn
T400 Bca d Butter, m)lk.

4he eclAT algothn to frd freuengt

ecAT ceqrivalente clas Tsons.foneton) i's a

Vestcal ata mining gontm in a vestial
toonsachion
ensets ehh'ciently. J4 represent counts
TID (Tean sactoo Ib) sels and
ung item
Homet g
heir foc ueny Veyt'ca fone4.
Datoset in to
Step:t- covet the
 TID set

Bread TIOO, Ty Do, 1 sv7700,

TR00, Tq Doy
Butter F TIDO, T200, T3D0, T4DD
1600 TRO0,T9004
Jam Tloo, TBoo
coke TD0, TYD0
TSOO,7600,
ilk 5T300,
1700, 7goo, T900t

Step- min suppoth

Find Fruent fternset with
Freuent sets Cupport 22):
singe Ttern
?6. Jam 9,cote: 2, m k6
* Brea di6,BuHer

too- iHern eyent Seh.

a. T900Y suppt
$tIDo,TUDo, 800,
,Buter)
2 (Grad euppot y
T00, T00,7MDo
(Beod, o IE):f T6oo,
(Bréad,Jam)}t00,70oY suppot
upy
(Bu Her, ilt) 7300, T600, T800, 7900
4
BuHey, Jaro)T|oo,T800 } ’ sup
+ C

+(nk, Jem) 4¬go0, 190}

3.Three -Hen totluent Sets

Buter, miI) {18o0, 1 q00}seuppot

*Bread,
={ 790o, TS00} Suppot1
(Bread, Buer, Jam)
Step3 Output fotluent patterny

fo uent fternset
Jam,lote, milk?
-tern sets f Bread, , Bu He
,Bulter). (Brcad,milk),(krrad,
3-item sets : fSread
(Bu Jam),
tte, ( Buteycoke)
Jam),(Bute,milkJ,

3-itcmset i {tBread ,Buter,mil E),

(Bod, Butt