= Sy = ] ASSTENMENT -09 8. Horshitha | | ——— 21165020039 DS-Albha | ly Principal Component Analysis ¢ Pca) : is Principat Component Adabiysis CPcA) technigate war introduced by the SNOHD EIST ESOy Teen tency Meme works wile the chta th higher spoceiiclsmared Ho data a | looney dimensioned apace, the variance of the doto tn the 0H the — condition that dimenstonar lowsey Aimenstonal space shoele? be maximum . pth main goor of Fiinctpad Component Analysts (PCA) is [40 yeduce the toneneenoliti, of a dotasct while | | Preserving the most fmportont patterns oY relations hipc| between the variables —_ystthout ang por roses of the forget Noxiohles . ~IN the context of deep leaning, PLA ts applied 0 redkuce Ae puber of features or dimentons in Pe input dat: =I deep learning 5 Chtosets may hove oa al Mumbev of feature, , which cam lead to computationak chatlengc, and overvfitting. PCA helps by tyanstorn tng dhe origena components. These components Capture the most important information in the data: Dato. Com pression ; Afkey PeA you com weptes emt Your data ea & Amasier number of Principat components ved ucéng the amount of memor and computodion veqywud i deep ead) Moclely » The,comprysioM carn (tad tp better general gabon Noise Reduction : PEA co absg help io teduceng tte Tn iD he chfa oust rg 00 the “FA ts ued =r enticing dimension, Algowtthr bv fracas Component Arolyrd, - v Step-|: CoSuslate te raeary Step-9. GYculele the covortance af oa maaix COVCH 1) Cov (%, Xz) | Ss OVC 7%) Cov G2, X54) J Step -3. Find eigen Volts for coxwmtane matitx Gtep-tys Cobculats etgen vectors oy Covoriance main Step -S: Corepute pied Ves by har Btep-b: Compute @rindpak —componen Exosryle: i = Given) the dota In the table Teduce the dimrensiany From a to 1 wsteg PCA olgorttiny. Feotute | CX) | EX, ExsAe Step ~ caeutating the prea nu Ly 4Q4 ‘BET > 82 2g eu) me (Axe | G Wu Fett aa ty Tyee BS ihe) Nouwd (kb us sepresent dhis dota tr Graph caf beesentat for mr 4 | af rwuth il er & = a G | | | | Er aS a o A a > xX Shep -2: Chelating the covariance matitx Ie xr, XV) Cov Oy, x2 COV Cx, Xj) fo’ Cay xy) leoy (Ris Hf) = 2 Cap = Ne Sy (Gin, = ) ae —e B eee @) Cov O%y%)) = += 2 0 mi) Ox 2) ner =, xIp-xi) Coe -%\ ev Cetyegs4) (aeejh-s Cia a5 CRBS") Wel | \ Ce uy ec Go) 4 Cs 4e0) \ y UNOVO 4 S041) uy? \ 5 Wy | ORCL X ayy OY tov O41) Xd Net oe %y) Ok X35 uy @ Gay 7 (OO eV RD CH-@+ 9) +018 @) ES ARSS Ree) GINSGESS) A (cy) Cas lo) CU SD4 (SLES) + EDC) ° + (1946 ATS - 5-8) , SE a Boe ay Sy = = Il er Pee eee vis Cov Ky) Xi) ET] ) cov Cx ¥)= Cou (kz) «) GU CXd, Xt) == é She = = CE ea Ca Z. Cre) Cae-%) miwllbags ae hero oei neg S (en-gsyre CUB SV CRS Je GY 4S] T ket cok eee ye CF yrs sy) 3 =A (Ges > ae (ES) 4 2072S + la as + 30-aS) > Slee: a 3 [cov G4) 203 “The foal covarianme wotttk s. fin a Gkep-d: Finderg the eegen Values of covariance matyrx “s s-az) =0 Wy NI ‘ ah oO Ss a i i W-k = ©) -U aa-r au-A) @a-d) — EN EW=0 dQ -14d - ABA 4 *- lal =o Ra =NGy = aN B22 121 =o a) m- B10 t20l20 -b t Jpevac piaelL ee > Qa ¢=204 AcDh the C37)" C4) CL) Boy (369- gog S65 Jesu ISee 236 161998 6 y Peo \ bY Fluac See See Upsuae & “Tye SUS 6A? ARG = 31-0476 9798648 GOT bMoKG 48 = i 280a7135 Sea z BOSS > 6-6) ww 61 Slep-¥s — Cateretate een wai Ve bel Qe Res een yy thes aan tele CU- JO, -— Wy =6 Cu 6 = UL aoe om wu 1U- & Di One Pe ea acbe r SES) Co %, aes sone it erpUtatfon of Canl IgE veclors Pore previous, | ‘ ep RS) wwe — Now thet ( by. On ie y i Oh i" tl WA Opeib | Ge= UY-Ay4 a Sup pose aa | Ole | Opslity-r) i fen Ve cto to = Mt i (Ur —unheneyey use want tb catercbbe POA, fixe t Coe have tp Constae the loague eigen vote then Pf Wwe WANE fo Conntdir oF cateutot. Neat Porn then take the ext ee value ( 40 according fo Step -2 To ind unit ett vectoy comput. ‘Re length bi P < length by be hy puta & J Ces WO [Corey cone ~ [Os Gee[raieace- ety [aen-200 = 1% TAU Unt Sarda vector wy piu -a/ itn | i” sees D-8303 {Lou 2 [wot etoye ai ude eso Subset ate the -fo coterie, onotha pea then Nodtor oh * Wat = [Go™rOV Sine nye C1466 ieee ae [or se. cia Jrs-eray 13- asig Unit eter vector. bo l “ : oi | ae al fv ill - [esr | PitAb: Computation of Fast pitretpal Componenes ave Ave UWen ag eyo 0. 5$13 | = 0° 8303 al = [ose -0-8209] | -UWe ane ee pinctpal compsnent Or 1S ctomple Qt = (SSB -0-R303') = (0-557 (4-2) — (orga, Gi-e.s)] ~ foss73 (<4) — 054303 (-a-s)] | = 39892 -2-0757S = Ste 305% =toe are colerntolling pincipohcemponents oy ond example ev = [o-sstd — 9.9203] ~ fo-S63 (8-8) — oga03%y -e5)) =(o-ss13(@) - D-8303 (-4-5)] ~—G0 +8303) ( curs) = 3138 We ore caluulocting prtrctpat compenent 2% olCoos 0% - 6 8304) (Co.SS73 (17-8) 9 Rod CS-85)) LossBos) - 6303 ( 45) O S813 (s) - 0-424 (-4-5) 2186S + 9.90608 S: Gas -\Oe ore eteutating (Vinekpal cemnenent foy urth ‘omponent = [o-ss3 -0€303) = {o-SSBE-8) 9 “S303, 8-5] = (PSST = oRRVlS-S)] = -0"S873 -U 54665 Te res Es Feotuy | xi oe 6%, ey | x u g (3 x | Xe CS me ST 1y PCA =UetoS4 | 213 63 | $6 Tas” as 390 © Siogutod Vou Decordpasttion : Bing Rog Valu Pecom position of O matitx & @q fo. ori zertion of a matty wo 3 rotiices — Mothumabialy) sud cn be exploshed a8 conmide, G malbuig val of order “mxnt The weteuie jn be decomosest OxLA= ov] UW pram A colin ovthogonag Chol roean, Ws columns aye eft veclows of APT) NO ¥ CAE LOVT VOLT ~ Go or) VY mx anc onthegonod (hat rnarns fe calternng re equ vectors of ATA) CATA = vOut apvT = vo" VT) po wan dicogona! modicur stenplea form: 0 = 9A? —Kingitton Valire Decompostdion ts sinrilay ty eéqun volun Necomposttten ; Abe only difference 1S We are ape liying unt eS Vetas —Tngleod of taking eigen vector dareciles, Ngorithio doy Stequiay Vos Decomposition: step-tk {ov es Sv We must decompse the metry | (aso) Step? we have 4p dined the tra.nepo se of ratue A HO order to And Sud Sep -3: Muttiply Ant ond let us AUppose | Step-l Find efoan values dr AAT Slee-Ss Find -~he etgen vectors and let Us Sppose etoen Vi, ve toarcer lo get Ve compcete, Veclors ve vad ete vectors etarn vyectors tnt Step-b: Vind vt ee qo tnd the Atagorot matrix we have to Calerctety that {he Toots of We eta Vous, ayrat feb we SuppOs:“ipe aeagonad vata fe Tpresemled a4 | & omy Tl eWay ot \ a) prep-8: ‘The dkoageral pentane’ AR! be uke “AreRLUe ete Oe slo Catentate Oy ard Up vabuor posed on the Sica matty? von bot Atcles, | A. vo" Av-uDV'V aAv= 00D Avis UT, AVE >Use equation A. opVv! 0 Wo = a AW) OS Ga AS ux said wwe can't take the etoen Vectors derectl 4 woe fave {0 Consfciy the unt agen Vecto ts U ard ba ep =20 i , & =e Saat Steps Combine Ur and Or t9 order fo get U Step-la: Fioadlp ve have YH piove A= ONYT Exounple: Find the strgullay vector decorrposttion of o ants lSa) \ tov cateutating SUD We mee clecom pose he matic cy Ns uot Yee hove cto fot We hangpone trettatie of matrin 4) Won Teg | he n \ al yA ead Sy wctliply AN" ANT « y=) fe Ho \ uy —) 4 | SUX UY SDE] cate TROY" (e-W & LER = T tet 4 Up VGE LAD 4-29" e \-s 18 Ts ra ae Br ral Ay EEN wtees of ACY | | Inat-ATl=o | Pua eRyt? ie na “> ie “J eS ee een ee (6S-AY U1-A) = Ca C3D=0 (US)UD ESA -lTAF ATWO AL= O =0 wos -BA A +A> —lor0ll =6 n> #2 **=O AO er see) : Gries eres =O)CA-BN 6)-D=0 ra ga y» Comperte etqan Ve Ltos wlvern ret AV = }\Vv A ATO= XV ded dK dCLOPSE (3 ] ie ca CaS RI Ce] 65% By =84 2 mere 1Y shy Brit ott to left hand wetflse words 6sy-81%- BY =o Sema oh aaa he” a) From ePher of equottoty ay ny eo] r yoko Qe ay = kv Ag T= Av ak VS Aapyiote ve] 6S 3 ( ae iW [a= ‘G) Batng atk WL tecral GOly =6> ay ALY oO Qt Ny-tyeo ee GPR BAK. g Wee KYO Rom ethic of the eybrteng = = oe , Ahi | Nie G | 8) Compre e€q0) Vectors tro the form of UN edo ve CoS vie Nea uy = ke i ee eae [soa J ee orca ; ee cole Vee =| e886) fe. we) g(a3e6 ke ehhWei A Ve * Ok Lp wake) - OG), preg | VT O° 8 (pr . y p oe (OP ge tp yy or ty Or %7 | Sie egret “fey 3, Dfagorot pradkike Aho! be Nox ral 2s Meatrize on C1 oi NZ Ome ie 41 a eae “ELT fom [2 uu? { an -0 +827 1 oe te! oO: ee BS88 | 3. (2Q oO OUF = (oe 029a WES Ge. 1 (AV) ih yu = | rl OG] \ ya } ore . Oey” eae “8s oll | Qo OL MW) = Ur 407 Urs oOu4 ] . “= OG” iss 0%9 Or 6 18 w= if sous) ~3-046 0+ 0% ne si MH pz vov' -U -T) _foroug 8°06 ome ea o+ (pl7 * Q ~\er0tF Use02rd (om a7 O-eq4 a mal _foeGlel -B-O%G “pery 0 UT i ot 0 +Ht Q.083 Eee 0+ 8ty = | ~DPASST? -3-S%WS69 OF] =~ >. (93 (24 =0 1268S 4 415728281 os 208977 + BS 265. | ig a eee |. 900 BQ [Ts A= opvt Hence proved,