Uniupervised mL. * clouihiahion cluskaing prim Recommdals “3 (ogith'c 2 kmcang KAN ° DGScan poe Memon = OT < hverorchi'c as Nove bye J er i ee Re bored ame nae nee ee GBM f Sui PA ten (cA SvD Aut ANA a ii Encl ing tr calbooat, Pcie Supe tied ML—> with, Reqeet ton —® Linco incor Regrai'on i a See] Muth ple polynomial eg fasion linear Reg retin Regression # (mt R) linees Reqrasson (S42) *= wher for= =e Raster of inter Regression _ a A : fie el caerg hrc | Yemnte Find value of tn and ¢ E Peer, yl ——> x.) E Pa (3,6) ——> a i Se inewelupedcs Vecvin . 6-4) Ra orn i212 m= “=H =9 Slope mcans wilh tery change rin aw Vole of % how muck’ bitter 10 ¥y 2¢ Py (2a ¥3) —y Cy, 2) ye mx +C gear CAC &-BEC= 6 i : ete tabard Vinal eg”. Linchon trol feloke % od for giccn value o} X We Can fad Cosrcapond ing wAlle of ay. BP L ndeped oeps lal alsa aL Pants bGytg t= t Gnxn th. 4) siemple fo einen Y i wostLenthe dola with linceu telohonslvip ges: 1) Nut suiloble for colo heving Non- incor tclokonship . et tyuue, Peake} ul Lie “cal Uaderkordi'ng 1 ¢ mi ca. ed hes GE Oe 1c purpore tah Aa bs bt filted line Cooelel) : for_lincop reg recon Ginga (omer 10 pictur. = (Mace (Ay , |) = os unductond with thr Ox. m23 422 J (i 2 lo 4 4 ue := : a = Now find ile foce-funchon — loss = [Uo-8 caer) FUe-t4) ap araayd Caeouay'] es 2 Wag pleter a6 2 doe ie co = teoson fo toke quar. »_jothok postive and eagehve : veluc_moy nel cancel each other E Minna Epis eylt ey. High Accuracy , hugh [oy value —~ low: Accuracy . minimum’ 16 Beh des at grocer fread 2 iu a mt zohoa refers +o diteraining best porometer duck i cee] ie {tho voit step down unkil (a tun bo j . qlobaf minimum by ikercohve (re ces a u c 4 n ' h | 1 ve eenaes = Rordom fleba) jath‘a) havnimum ee Cm3(4)_ ta mochine earning fie [om pop = LO m 6 a7 |, Inkerce Be = learning rate: Fal "5 om ogartude of chonge thol-you 7 ad in poromcter dia therohon re <_porkial er obive of Joss funchian w-r-t m 7 Reni © Hee tb ||{L (gees, wu'gh ond bray ze ys wx+t b | an : T (weaqhh) aa) | k= indepen deal Vartable - | Y= dependect Van‘able ae | Ww = Wagh o | we Te we fi pq in (ors fanchon W- 6 b - bras ey Pat Pe fr I eee ions change in hosiFun Hoa 2 6 Po bie b = ad bar, rnin Leo i) swing porometer In_an ophmiza lon olg orth thot defermine chep size ct cock Meehn while moving ctword minimum of low funckon . divs =a = UNNI) v=o LN “a 5 se Os SN dh 5 = 4 ! redichon & a MLR Not qty tats ts Seder nese ee | eS Ape OX tae Te tagxat + anin tbh + & SS a eee f predicled value 4 Ackuof Creo Voluc -Near Regression ) fon th line or Tegrecsian . oul put-. i es Gice relahon | Feahyr and dor get output gokve corelotion ie .o Non-lineav(x) Nok ole Ueobuce 4 vetorg ct) linear Vv Vine ow Y abli ty multinle | : do check tats; ScaHer plak-: feoture 9 Ve torgch) . : oh. Ubollinc ELVEN crea ntolverc feoture showd hy independent ov should nok Bei os oe ee sueshst )Tr h 4 imo a iN | vif CYon'ance InAloh'on fee ec = | | lal it aveund | ther feoturt deal hove the issue | Lemd i} J} v1 ov more than Another thot | iperbcular fechuri hove moudhcollincrsly jusue and | need fy remove it I a}.| anuthes method fe Wy find out po relohemubehaec n al th. feofura _( Heolmap ) - | - Normals Ags pow i rained ceed areele) I | d \ tandard Moma, | | disbibuhon means nmarimum error shout Loround mean = I ho » tn check pb: kernel deauhy estimakon [ kele) | oan. Qi Or = agente bee) (Nol déciroble Tal docheck:- Statta plot ( “ypred revue) ) — - oe) No a couclekon of buen | if wep Raia Auto corres honi Negehve Aotncorloker : i Nok dee Treble) € decivable}— ie check pil. plot (1atduel ) 6 ewerity predi‘chon Tafecnce High vi Va Medium a Bg ans wee wv Ai Yvl] r non ln emg relahy p. Bot Qim+ Bix i fegression Ap J Fur de Wes Miss, : ‘ cole a ned dake tur train “f Ran Ehiead palipakoi? fecha toy 4 cabrecf rey y= pot pyocd Gao eR aseewl f: h r Now how we will tno the efec value IL afte Aine this fs hy perp momelon[vasiall aes rh ee ul, 3 Lele ou Zero mean Ma elas variance ey thal fous ten “ pallincarify (high carrele eh'on ) amorg ! Vins indped eh vonable. * x l ee wi uth be weltcasr ah toe — k. nal n lel urdustand avth pablem k pI e i | alesis iS |e ie Li Eilts vs [a-2]¢9| 4-0] 2] us| 16 Awe , Sava ese ze 61-2 eases oa lee 2 3-8 4b 4 Ee xXY=siy-g =X = 140 a 6-4 a0-| 4 4 qe 26 le n> I fe 56 1h _(numbhi cof deko poin hy’ b Be 81-6 36 ‘ alapoiek) te ote te gee] D4 cre 34g t4o- JL math restr”SIR APH NYE (2%) = = —O-srgs4|( ge) Tinea. equabon eels outed plug volue . im aie math. = ny 2 a:bIX - 6-83 lel tak it. for 2 y | Yack 2 a-4l (%)- 0-83 2 3:94 3-8 gy: o-4i (S)-d-&3 = td ee cs sa-4i(t)- 0-3 = [6-04 ie “#reoding dolo from ésV / * pd-reod sv (‘tron cev') +t dehining the variobles = db Lite!) plist O Be del Vy]. tolclt)pate: / / (Coe > dt adding -thi andor dermn A ess . . tad cvnta a 4 pesformming reqrasion aad fing model ———— Feud} = Sem+O1S* Geo SS # rok Summary tobe i (raull- umerany-t ) y=}overgi Hing ~ncons cto recluce Slope ae pai cua th Nin which ie ee a 7 a Nok en) Now We Raves Cun vey our mode ty choose LR on Sy pry tokPage re (he ae ——O iI ST nen add peasy before we ute teddies ——tuhich-waill-add help this teem for Hs ial term E | nutaing elecrvews Ors hyncepatarnctey | Raiduel f m= slope. | Now we will co | for FE ome) with eqe © E | Since Ine tb posssn (ole obiec\icr . dheaugh both points (8:3- ee F-be + (0-q)~ perfecHy trols why Hl eben beruecs = (ors Crs co-g)> | 1 (De Bagi a =] 2-03 | Tt ees = inte f _ _ Ee eee Hae nud line | > ( mitdinn*) on A ‘fo 2% (etm, SEme | Aire we ore doing sguar (*) a} Ha beni we yaled-tk 10 som 9b. regular}nebion +i Owd ¢ Tu on ath y vance hon gf 2 lg ¢ x i ee = Gra ye Ae [ a | = Pe in ES italy uy agi eE yw! 4 In ridge Fegresston £00 Gry voluc ofA thee wee eloay, tom« voluc Ay anal input feature buf in lose EF yuu conkneusly Tacreaye vel cee Awad (onan pein | you will zee yalue for som. of Cong nene, hs -an_oot imparten} jo hea we unknowing doing feat Schechter omd ih i odvorleq: of Lasto +110 wh: ing 00 high olmenst¢mal cola Lend) Urry feabure ore nol tmp { pcb ,We le V-) core -0-0} ( a es esa levied lined? 2 0 wally or you Mrrcooediak eee Laval uulone detreare ee doword Zero % pwalize no on Gia cod ANG Ube i Peer know if A inerecu /Ovec piling pdereere 7 tr (Acreaye ta pian Ce) High Rios High Vonane mab i Und fit over KH __)Under friby Over Hy SS | ho ovid the ove Bros -Vortance Tradentt- Mi ts_importont—to undeshond pred Hor etrore Lhias and Variance) wher 1} comes do acurocy Loin omy moe PA gan" 7) { _ thee 4 frodoft behaeen model's obilihy do Writ cni-aela sere cnulvaq one tahick jy rebtredf b : f ‘o or scl ching volus of peguloxzchan “Conon pope wualieae ng af these error: a help 4 th ng and sede Hing st olose uhile aintPol atonal or oe lenge ee (n_brasaing au Wal a4 feat fe iat eT tay iin bins kl, ee me 4 peiom bee 1 mupeneclole bag hak vonana Aa Ta Joot lke drilougI —— Blas sAdsiereries trade offi-_s] tre algevithon ies Jimple Chypothats wit. bincor eq*) then i hig, bras ond low von'ance ond thus + ir 5 - a a i Lequehen | then th baer Ba on high Vanonc ond | i F | Gail, niefe lo tes contd han ob. ! - aighe ee gi ce wilt ee eh oth t etudee both fac _condihin ee ab _frodect} oy bia —Vorvance c _hredeoff- Lf A) | € Flect vt Regulerizchion “er los furikon (015 fun chon! — 12 fen ye fe 2 ) Ir i ecoecm hgwilord on A) volue fomibr bree value + -jf we ore treiniwig bids been! moduli ER, OF ae bb know which eet perform betten and whiu, _petomsbus: ore beHer Jou Ainchon py useful, Clash Nel oa hsieu.) t ibis Corbiachon oth reg aaa hoe techmnt ghee Pall cl hat —— Begue = Zeon | Aimee tlle n Saaeant \ | peralty which iiiepased an pore"es * fap pray a msc} | din a ‘an p ad O ona & Cidge tarso Redge RIOGE “06 = 5Rry. gedae QR svvy Loss. Ridd qo to ul an a ox'xch'on pic oaa gain to223 errr tard year : phe weoor overall) wud We rq undersbanoll' ghoul 2 teven tp _my olato lung Tern hal hanpsa (MatMat my 3 2 2 Fans Avt tat toyed ~The ; \o. D8, feb: Int tint toe + -1 + Tne (o. aaa NeaNes ctesey) - - « * Ning? Yeo Tem _ Sim,O) Simple Exponhal o> Liv applicabl -Pothose hime _ Soe Wethere te no Lvendbno Ucaralaa |i ity. arin Holt wrela'r model - (tiple éxponha Lik iS aes well whether is both bead as well as ae Risheae y puede noun act (savideaiem lnowitica ot seosonol: fy then hell wintur ts hel) Sromple Cocpor h'al He . Dy O. Da Digg Daao Dao} tomy sat iad she ve aba by sigs 41 toy, Today» (orc!) webal will be the cuit woy to _proshact ire fom : Mero? ee ee ee tito r uhihg Avg in tutte big: |. scteaaleal 2 % Loo his nok good encthod aince temp von'es Q lot fhe oun cor Summer fo iatintey ] ’ ane ther woy 1s Whahever Ce cfod ays emp if UI same for-f> movro@ . ee fe amaliqq (ree local Av erode ab Day 4 je 24 will NY cabin ohon t ae yseled u'r bovos futrage ond ted Lb i j ie 7S 4 a Peak Lah ey aq fh Liapt % +144 Spi benpoeee 0 a Liqp + pet ee ce ! Pobigg +1 ctiep ® fin © a ——=— = A 2si rearrong thu venue ne e \ / \ \ ( Page , | ia Cee Che , ele Laon 2 Lobe ag ts tag Faw og OO | 4 ° be from Cure loporel wer'ghbeqeo pul rede exporhadly wheucas 1 sionpls average tary cocffur Pee Let undeuland aith shot Example. cE 1 a a er & 4 1 Jo a _$1 42 Li lala, —+ Lele. UetS a0) set JA Mt LatSy 5 beet 2 3-4" rise pis ni RS Lv L Pe ea Gmc =| 1-6 wy ie ro 30st, Les 2-12 Ge aed yer whieh ir not true.2 Ling + Yuse ee oo) +, Vivo trend i's cada, ln ea Geile ety) =O _ 42 (tal, G.-u) 4 LeBaron ple 2. ie ig ee vblum » - Fen pl bxponbial 2 Double Erepenhaf .c Dates / / \ | (Cn __)) ; — ri in = ae eile jest oF ee Hf = ey ELOO| tes ae 2 jae y ober ase ( I: 2:39 23-4603 4 6494 Wouter will ealaatals wi Deubl eprint PDL A ya eae Ve : to Le la by bye et Sc Le | : =, ty Ta TH ot les Cutt) + ae) Soe fala oe] z Loh wae afieel iGo ely (s-1)=otor Ona __ L Bo cee leet ls.) — ; Es er : : 1 (pete ajtd (4%) alee act 3 ‘sa = opted Io _ = 29d E aries Means ne (ta=ta) —— gene ie wae = 12) = oring + 0-437 92h aeYz lesser) +, (9 4684-2: 396) pee? D229 Meg +f oer (10924) eon ee fo Su CES “2p pial + Y(t +Ty) + AGT al ecto teer is wt) Fe Soa iesy bag pcre Y - Bale: (bg -Ly] : Yfo-seoi)el, (a euaq— 20009) is O-4i4os t os (11494 = o-4i4os 4+0-R4Er 2 Looag at te TTY Treo ic prircase OY OSE o-gr Biase| / ol. pat dtm ¢ coies dola. - Hcy iw Ys, Yaoo E eee | Eimer ieg Lroa a vend | Ste the Ta Thos r ancl by Sj 844 $200 y for 200 Months ——» 16 yw --F mensth for sco. on | by = periodic hance of |e Mundy lek fiat comes i job oe ep on bial pablen hap ju b m Hom t | Mi-81, ¥o-S2,Ya-Sa ete EN 500 aS ap oe Y (hing sariiael ae fr-(ses Sexe) 6- ; {Note Aiaciyoo0> Yrou= Sree Zs pate we (tag \t (hoo =Liaa) Lay Soro 2 fy, {Seoo=ty tip (Sree) z Fem@mcukcon del dulanoo to qc} bree rt subsbeact si90 Grom Yavo fr _ NOW $706 wt com do Naoo - Looe |) Em ce VY, (St0-|0) + VYo.CYnos Lave) ee)A ne ie on up hy g _ || for Looe we need s200 for s200 wenend Lavo q . if be , * Ai ins fay “a a a |_tne_eg™@_ Sr0u =Sigp Tae Is (4aq+Tiaa) ‘la ae = Ser) ——— cay. (trae) + Y, (bree -bt44) i Wage = Vy (Seo =I) mG (Eynoe =Lra) : Ra when we cam whol shourd be __ app! ae! futurc_weedschion : lost possible local Avg = ih | all future predichon | hynep aromela changed free laud pacible beat Ave k | pyonal i trend ard! local Avetag «+ been +MindeW Reved according 3)ARM Basieadly fen ou thees_madel s-— si “a Tenia) bh) AREA [tama | er Tae] [va] dole shautd be sener_ncsels the ie Here delo mul be stohonary it nncons clisha bubon a nme] po Mmrban rn An tag ( Stobonory) 7 (Note takoren } :‘ © ey = 4 fo BIDE Mte AM eat) eal Oe) Ye = AIMt-1 t% Ve-a fda Ve-gt ete . AQ How to scleck Bek poaibls model. O@1, AR. HR3 | ret sehr dard on fut dota accurowy whichever will _| pecforan good wilt Anolis thal. ¢ ee Yt Mal Sys Es Mt 2 Pa & Et Bee hee-) SipOuipne ey MAC) a - | it lal year (onl TEESf Dato: / / > | Modi bored freccashing . — ~ AR a MA al fram (A R+ mA) TMA (which ia coms by ARM e ilsel b) ~~ = : I mo | LAST pee => future predichon are comp) ulely depcao] on pad: dota il te eqrass KX 6 IC. E a ee E ea predichon IL = Seat t ere arts) ‘s ibiFAS depend on on L ae ela ee gett tt at teat zetia Ss * la yolkerday Jeu ie J yelrd ay -alo nosn ee Aa ooh D1 thro A ' Jo PRIMA Bie Aince they are enced te a on ‘ f L lon ; an MA nein upon external fan. EY ee ae ey yy OY. 49 Conroe ey | 6 |e ‘ae oly depe ding on [eat tiene peinploxevale Yer viet =! tohrot-at eté er Ma(3) 2 pense toe tele Eee aet pte. 3 ene cxo tly eee Be model only difference un intlead of previous dola hee extemal foctors are Jotuated: MAL: il iv combinahon of ¢rhand M& mode ye AemA C1. 1) a eR (Yer F maci) | Po fremp (cp, 4) = #@(p) t mac 4) 4 Baye Ale ae be Bodie Ce iF 8) et-1 72 Arak ail be AR (1) Me ieee py ken) ft Oe et E_. Se ee Cat eer PemoroMidpdlpowiblepar woke e671 3c mm ‘ - LW ARIMA dounf Cate about dela stan paany T > Not. lk na bulicxtenston of Rema 1M A ae ‘ ive eg tole proving avcte qe ; f cformed dolo fom oxigi nal dota follows ARmA a original dela will mut foljod ARIMm iat fs Yoo Ya ty secee * EK ; db Hrenvfared doty + r . Zo aw Lourie eee a eS i ZL we ayes i ie z PemeemocieraiteD FYAIMA,pate: // | \ > SS — enopeics ———— for ey. sy anf polance o| Gory meoth = Io fP aes get NGS Vy Bh nina 9 ny (today) : hes moni i addes) tw ery manfntl die ane) i ee ee ER) Tae bs i : Zy he Ly Zs Zt a fallseeoee ey Ny \ thi puoly l astucled dala Ln Jy Ve | if his food Pema the pare Bree yee ase of dele will follow AEA #emA how two Companeed Ce 4) whoeagy? AeT MA hos dee Dorseatizl (pgs dyNo = aaa ckepeeds on stefienqy past Stohio pari Extenal fecker chohionen —_facken ae Noneed since bom ipeime : tronforming -the dota. | £ oh, < 3 ees iad Forecasting , dela Bore ae Pmoail Qored | — Meise) | | s are ARMA) ARIMA — Drable Hale wirly, I t a ele buts f tihen we Qt nok sure - a No_ 4 trend & | dive cHy qs fox RRMA gre i q v 5 me ae henever wwedonk rectal We ich eee Pea a aigin : ———— Loe eee =a juoce Ercor Mare Score. i» Meon squad @rror. Adual Predichon Ne ve aed Na+ 4 - neers Sate E | an u P : BGEE Ges Ye ya tare ieee.) f 1 MAPE Score. (Mean Abcolule Pecerdage ee) lel Compare 2 scenon'of__ EMSE Gnd MPPE Consf : Classittcation, Algarithen Pade igh's Re Btuion 2) SVM ; kun. | Deciton Tree 5) Random forest fearon! Pi an Sigmerd chon S08 r va oe = = Pht otk ee bobilihy thot (y=1) ae toys toy Pls: obobility of y being 1 | feo rule for givin Volue of x } ! ; (if witt be in thor me; Numba of feoture inthe / -a5 ae compl a tlobanihip 12 dln oe do owtlien ~ 4) Need to loge dafosets ae Moth Behind (oquibe Regression : a =e ° g q F LY. ° o i ! \ Assume “Ze sx t lo ne | . War On Ke ean Bez 514) teZz value t's lor, ny ye © y Ces Set Tr per = Dee 1c funchion A cost function for_logishic Regression ; 4 g ncion Lure BG, aw f Seach ice feqression ~ a” a fi =Yp« al) — and thy funchon we wit get many ) minima head 2} qething alebal minim 3 I : alt r f wa i 4 AlW: fat = _ ae = ie (Y) 2 (Gl. bal rm inion) () Cross cebopy locs furchion (ox) log ore vi { BY) ==(4leg9 + (1-¥) toq(1-F)) —y omit 7 ae bobilty wuld be Uenbinuou)——— uh iy: 7 es -l tog ¥ +0171) ‘og (I- -¥))- Ja me) ee eee i A iceeas alway ee joe es lows hee voli ee cree yshould b be ver large Chom 0 to 1) if th tr | the = ogy with be ver 4 large oo an ee small pumbe, carat 1 Ba ae 4e8 ee Vat eee ere 2 = pclealils -y) Bu «sce Ae oe ppt he see tmall_dg area atl 1=¥) w)) Jorge dita leq (t= y ) will be large maak tas. CEO Lc mal), number fon To YanviEe Safest ese (5 notning abe Mneon overage of [ose a “| Loss Function (1) anit cnn cu _stinale are set ay Compored 4p cost funch'on (x) whieh deols a a ‘penalty ox nurhes of reining | max ( 2 -} such Abel wu) ae i wtx, +b g + wr th >» emake etfs seme —____— —— mar Linsheod of stag fe woe com_el oi r — min (Ee) eR =e 7 = a : ae ae ae = (we encode fo oar | ¢ = Numbe: of_exar Jom @rror fo avoid ~~ oar rae Aae.g attuch £ i reo aark OTT he gid andqrain ond will bed fy fest Joka) kernels! in SVM: Generally funchon of the Kernel is +o tronsform the training ot of dota 50 thot non-linear olecrston surface con_he transform, doo hincas equation tn higher numba of climension Space ik return the tag product heboeen two Cl in stand ord feoture dimension | Pijpebalevoen Keine” 2), polynomial, Rodial Bans funcion (obf) sigmoid : LinearOM ee ee Oe eee pari CCU Gg oll Oubs Ug tied il Que aial Let you fry to plot on thin 1D Cine. f =e. —-- #—@—e— 60 © 619 «6 9 1 we can dee None of Hoe [ene Lowld cperale the fina class pufectty.. Jn “s hy weodd another teelure which 'e funch’on of >¢ fee aie | Ke amt Oe oo ny - yf sjantt| eee | weal | : + 4 bf hae ¢ ae ey : aa | Now tres ley - re | i ml | Spuable ~ i \ ‘ee fla, — faa = i fe e7 . | b5| 2) rachiad baric fundion, (nbf keraed) eat K (uis%.) 2 exp Copter = xa IE) eon | A). Sigmoid function c E a - Rou anion eee | 1 . : || 7 Se 6° 6 a co a s. © iY Se’ ; Cm e & 6 - oe coe 8 6 je ° @ Rod Morgiing. Soft Morgsne —fHte Medel is giving 1 pete J ft | too accuracy f function J for svn clossifien "Hinge lore! used on loss | function iki one of the tyne of |oct-funch'on Mainly used Maximum margine (louthrat'on niode| ane dss Lerot jor of es a margin br distance from | mien i orginc hom dlectiteg eau _penalhy tb th, a | Dovmdary ix nuh |orrgt enough _4 Le mane (ov -% (wT x +6]) O- for correch classi reohin | — tr wrong classifica y a "imei Mee peel ani ie ct= (=) Le Gi- © C1) os bt | pee Say = q both are hea loss vole } Now fal For correct daitftcot'on Yiel ee Vast Pes Co- (tO) (b-(-t) CU) cal ol Si Fz et aol volus Up! Se| KN IN = K-Neorert Natghbor me bbrevahion kw tonds far _"' = Re ls ten Lif ts iunpenarecl machine learni , | Bee leg aoa 0s well as _regraitan pve b ea | lever. textias yo l the mimbes ofecara! sctgbhor do anu vaknowa } = i Hy fo. binary clas wifchcn tame paceiple for muthiela- wuihcokion problem | shin quactal ik isodd pumbey | fundersbome with some eaxicomale | a A : ¥ ° Ey Now we hove-fo ind nud i m : o @ , . added &) | + o 6 o ie otscl: ckomce | enre oo) tun 3-1) Now bosd on ; A De Bee Sse se the to pak raiae bm sorbed 4 __ a re SM) <7 Sener fda taka! Ip. a4 a Now cf will aut! na claw Ee ay ore aah (asa aa of reser (Dud eee sewage Bock. 3 —— | eee | a | lede bamce:= ce al ere ae je ete) CMV) i by -Mochottan avilance ee a mM. distances X+%) . a pis ts for clombbrcovan problem i | as ler see far fegreasion poblem “eae oul pul tr | | Contin uous dota . eg price of house _ >) thissimilorto dassificoh'on onl val Pea to last sk here we were comidering mot Freqtcn class here fyork kk number of de pel (hora k') we celcular thule meon , theless HE Llronbebon } Nok opplicoble ta_hugedolout tinw roe oll 0 bof *| Sensthue +o outliersDeston Tree. = = 4 —— ee nee ee BED SS Model Cis a bebitee atiticohon & | Regremion. : SIGS! “Dutlel Decttoa Alooes ateockikeps — _) Goris of tree bord model sts i Steels, ton dos i Beomples en ree ae Can AS a tor be Ah classiticohion A Regrayi fon A intuprete No need for Normolizohon gr snaling Nuk sensekve qo ollie ys eT low i Enbepy and Gini: tenpur ty ore invers ly gassiteral each other aa 1M F Shea ale oo @ o. 6° Cuee in aunnO) 0 6 mote © oe 0° %e C= numba of dane Pi = Probobi lity oF ith claus| Info gain (41 f) 2 Enbony ¢ 7) => as | Leet beaters Gini mpc hy t_ thir meray of icmpun hy at Node Oiler ce dectyron tree ty sad to he pure if al the dolo point On accurately Lepecal ed info dlittercs} claw. tt meojure ty Likelihood thal randomly Jelecled dala Epon woud be incorrectly classi free hy speulre Node. | formule: 1- ( ty" t Pay Dy > probobilily of lacs y Px = probability of clay N , a it allel| Decision Tree for regression ; bbe oder stand with -bxomple oe, ee Os ease oe. | 3 ___No cae 21 | | 4 Wrong prediction s blue blue : ale Red blue — > W tong preduh’on Lf blue blue Now [el vedefstand my dishtbu kon of fous - | l earl iez focus Nwwo four _ypdoked focus a y/o Ola "74 6 sli fp 2K Stee 4fle |2 ig ek Vig M6 of fz —>r% sg alle is tfz osx Yi Vie a aR HE Alle ze fF ets Sos ag! 7: _\/le : il eG Z>| § © ince 2446 Gere wras predichon need [onprave > the focw. ond rat 1,816, F art Corre} wil — reduce the foius or wi con ole called it ay {i aie OPE ee UMN ie Ay a : a incurvech precich'an amc veduur hy 2 i i“+ jum of updold focus should be net qreter than 4 pine new focus wa !&/ry etl tac hey do , = | Novena) fae eid =a ee do so we diwd: New focus by (6/pu4 and we get i dot | new _updoled focus whose Jum 1 equo| Eo : Hous whic hanes paint gel higher chews | eee ee next _mode|_and rare point with iL jority in alec ill _gel_lesces poi nib —_______— Towed] [model | fede) eee eae t A ; | A+ [a aes I AG Y, % L sea a a weighkr n. yo we will treale pe dataset Ane eek aw | cd pr wughl recd ¥ a ik will _repecke one Kone f | Atac Fiak poral =| Pa : ae il el pod wh Buck nonge t 1 ra 00626 O- 0:06 ola 6 our 0-0 6ry= 0 Bing (big buck ck) HEN s 00 bre O31 0° 3FC 14 4 ore ots = 0620 (big buckch) Si De 0061 os boy — 0:68 FY aie 2 o-47 0 Gear — 0.9.94) (big buckeb) | O+ 064 x Diadivoesi Ine |S SOSA Se je ama hig qe Y hewt Oa uj getting fous ir more while Jempling with replaccmenk,( -— \ (. Page ) = —~ for this cxomp|y wt consider cenilo ri 2. for j imcrcoing or celuciny ths fours ud how we deu'de phic \Constont . 3 | | tala Sere i whee + © 2-18 i + learning rale (il coud be Fron 0 fot) : el Logi. teEnwor ee erro fox éx Ciror = (« (oy. —¥ o-| k: yl t-o-l ye (ee) > f- log (9) k= logvg E Acs ea eee or_dinds by @ bo_tncrease oc detreas pwn hyo teybead . 2 for toy. Srror we would aa | Cun mony! = shot iF Divdd dala into froin ond bet ath creolr weok leormen ( stomp) a; Colewoke error , find mushplfeoh'on fortor , volue of kK AL > updote the fows ond updoleth: dota creat «nao model - _ with updoted dala, Sit repeal skp Dand 4 number of this tepeotolron equl — do numby_» 6°) Combine mades i ving :| \( Page Y } \ Grocltent Booting. 8- tel understand: with exomp|e i siesd of fegiaion problem which ir Loy bund istand 96 Gem_mpare clawishicohion “problem. “a | ig cg po Jadlorg (hea) — es go g 3 = a H poo + 4 - =a + 116 6 & — | lae 4 e a go Fe a = le will create gtocier! Boakng Mode of 2 base mod, } feptalnra [tml ere | (mem) (0) (eT) loupul of Fink model ic my will olwayr he maces of- awlpul ov forqel culumn nf dal oset aa ee eeheveailh iy Fey —s pred 4 ced L_predot en predo te pred 2 u : J en LD ree a { I i 4.0 — y 8 foo | 3% 4 API = 0° pees OO lo | ¢ g WB Br Be 8B 1-32) = eel 9 6 pe he OR (08) go | = 3 eG -t-% _-1-8 = 6 | ee In order fo fei) the error dance by My to Ma we will — coltwtok Joss funch'or of M4 actual - predicted ie Psedo-residuel a —— Now we will trun modd & Which «i DT for whith ae a fopul will be some Cig, Ggpe) bul putpu will be ~~ ed of oy pedo fa duef .end not Jolory . ;J [nfcrences= Here we ort aking. do modeltdo fetus eres mfitake or error te model | doings. Decision Tree. i model a ee et _ ! =k | a | [aa | fi Aebiee how would i predi'ch iP 1 would have Bm of } Lonly too bored mode. { lel coltutote for sludeol 4 ( 404 &) > tee + (= 8) 2 yee =e 29 Studen| 2 (100 ,#) 2 Att (cog) = 4eveoe = 4 student 3 = g [ctudeak 4 = 6 | student = 3. ee which tall equl tn adual 4arqel column ,tous thic pocalelean_of oveifitting. like th wilt good for iv fthol'y why in predichon we odd leorning rate. predich on som ct Lo ¥ ma Genera 4 (a= o-| Now itudenl 4 42 + (ol) (<8) = beet (0-18) SAC cosEee asl oniih pnts dele Lalo ing abd Hola Now we will calculole mode 3 fir which werhowd know — the performance of mode {| 2 for thal ogair Wwe Ol] _toltutok prede residual - ~ ai, Ie pedo= test « oduef ~ predicted by md hubs. | ee 2 One) z + o-(4e +01 X (-18)) = 3-(424+o-ls)) 2) OB Chrr = 0-1 2) See enme St 6, ‘2 | for shuden 5] fe and ct will be pactloteidualy ta wiiton_cleorly ohscue—oftec ach thf of modef Wintel! ties INCL er 2 Now we will creole mode] 3 which Ir oqarn DT. Whore input col Will be 14 amd _egoo hut torq ef will predovotduaf ie mody 3 combinely prechic “mt Oe pret ht | and _mode| 2 Model 8 tg <2 108 | mz -om] [ave \-o8 | S| Now well Find predichion 3. —— 7 —— r ‘a_—s oo _y-peed chon h y—Gam will be atom A ——_ F my beitms + o-lx ma i Jel calewlate shilert * ( 607 4-4). > Toe a - eh. a det o-i(-1-9) + ot (=): 6) = Ss = 4 bel giSson se easy =} Carlo it wouta be 4°6> Dow ibis 4458 Diffeinse hehnecn Addo boosl sls Gradical Boosh | Ada boost l fo im Gradion! Boost } © | Hee we use drasion Gump: In GBM Max aioe: | COT whos: tdipth 14 ay leof node coud he foro hean mor [eePand: cowed | & fo 32 [ae model a Cm ma-- mea) Were mtu pi learning paorihzed hard on thd] pale luacepl which te ar weights J one focal ths med ds —— wpm twWome TIM ———— se = tLGro dice Goerking fpr claserfs coher prob! mn o— Love Mow'ed like popeorn hoe fovonte ler - yuL on Ih ~ Aluc— - yes |. a ag nyt No l ie _ oO | ver '9 _ Red _ No _ Noo 82 green No fy blue Ver whee we use qroclieal Aon for “clasificohon the Hh __inshtol _prechich'on tor uu individual ts +h Lo og(oddi Codd ¢ For dorgef 3 t poeple [ove the movin) - + leg Coal 4 °C don F love fas move) ee = to = of (nunde cd volue) oe faite pre Tee a el o Taub Like wii nial Reg reasion nth. causal wou eran sis eatin hon ta to ‘toavl bl po a aint iL wily, 9. — ° In _prockce pevolee/ten sof the rod mum number of leover fo be befreen ard It dn GQ fer peqrerston 0 [eof with single reidua| hed ot pul vo ue equi Ay thal rerdual. In Con hea} wher we usr 69-4 clauificohion dv srtuoh'on 19 hte “_Wwenpleg . Pris te becouse predichon one tn_Feron of tog Codd) and Veoh is dived from _prohohi hy to we core bp ea badd prem 4ogetha hq el_newd. : odds) —predichion Guitheal some sol op predichén =8- For! ouput of leaf te _ = _| = Residual - : ZI Prev: Probobili ty; X (I-prey prabobility)) _ _ - nO a - oF x( [- 0-F) 7 ap far oubpul of leak deg ent og) (ut K(l- oF) HOF (ov t) o- Dutpuk volus 4° leof 9. - - aes 0-2 +0-3 $03 7 [COFCO BECO * (1-0 FH CO-F RC 1-0 F))] - Mow use on ready to updok bur dich | - . predi¢ ons Combining the inthe) Teel oil, nastreée-Hoe penton, —____ avernale 2 love Residual predicted Ruidual. Gler mote Blue’ |) yey’! 0-2 ORG O°! qrew yates O28 or Oita blue No -o-F or aloes Ref __No =o o-| = b+] No gay green sydl.e) Piigeg og 4: | No 14 blue ya oS 0-9 o:} thie nus rected probobih’ ka are wont Pram hel L thal: why we avid lot of free and nokN ow) ra Hne baby pbs tatdual wwe will Now vee pers — [Age xc | A - fasts oO lor _| (2) == =a ol, -e-1, 0-1, 0-1] ; ay 0) ow pul u{ teof gid: > 2 Revdual —Land pos) E [pee pnb x1 ~_prsy poreboby lity ) 2 O'£ ' 24 be eC o's) —_pulpul of let [oh 5 oe sh : a a os¥(t-os) oo Quipudy of cok g > __2It CONF Ce 400) {14.526 pean) (0-4 v(tro-4)) 4 (0-4 x (tmo-g))b — E (0-9¥C1 = 0-4) ) F(0-ax (10°49) fo = oF ml Now w we hove col leulot _+hu_tree we ton to Lowe vedone gu for | the putp i vo aluer py if with evauthing else te al mh 7 ba procs, esas veki| we eho 2 ve_trodle Hae __mou'mum number of bees Jpea. cof o Or the residue oe rope donald f. 4 twee need 4a dlassity a nw pcos a ® somonl— who love or rot [ove the movieAnt lect + o-v¥ wd +0-9 & PP or tod e| _ tao del model $(ok XI) 4 (o> BXO-S) 242g t 5 — isnt pala li fy Hal ts individ ef luxe the monet ‘ - ARMS Biles haved ke e ore “ dy. estas Fy + Lhe will love +KGBoost for Reqrersion . a tye vole i __ dmg woh ae Vole (ndicaig Formtul. Ne tt 30 yy - — Drug Dosoge _ - ioth'a] predichion ir_o-c whealher il te xcbooct — —____clowifcohan Ou regraston a “Ravdual > obscved - predicted : eb burd tek pt with reridual the free cfod with - single leof whee all raidual will storef. los, boy HE, PE tsLleaf tye will colvubele sienilarihy score few raid uf —_ : ——— Bioleridy leer = [gum 0 Reridue] - _ Numbo s Rauidual +» ~ hee Ps? regulanzak‘on perametor. Assume APO 4 - os [-10 SLE rt EOF (-#s) | a Ato _ - 4) 2 16 a _ = uy 4 Now guchoo iy whethee b+ not we do belter fob of clustering ject os Yeatd exalic if we sph} them —- inte Foto groups135 fist for theshold we will comider two operohion with lowest dosoqes ond date their axerag efor SSLetf +68 Qiqhk - Roof ss i oe Gain We Cdosaqe 7 20-¢) & KGair ( Doseqe PIE) __eJ New we will shift Phrethotd Vole to go. — — . ——|__ ree e a - : fOr by EY Hs “soe 403 [ates S66, 4-¢ | so5 She - — Sse = [ato +6 loc +6-F4 4-5 - _ i to _ _ _—s—ié remove [weil hranck % Well a rool ond whols dree wound gon and t} woud be exbreme pruciog - a thy udid for peo :Now we will repeof all ths ukep for P= | ; atc regulon'zohon porom eter which mewn iF - : i" cto indimduol_ intended fo reduce the predichion sack observohion _ es er ry Fae - ic ae ae | : 6% [i nonenieeat - _ Tees wel _ Jtore ab roof— cos tee tt St ea ees sion Spel Hr - - : tT a a 3: — ” ree “ce Jef 2 (cles)t + so _ bb ype pestis rte EQ] 2 Vos ss. pght = Binion hy score while Re | arc vey | les than simon ly Azo a the omourt of dAccreose ix Jere while : jowers) y proper] 4 the number of coidusl In the nede- ; Gain: 651A tereight ~ ss Root ee ola fiw 6 7 Bereo Gains [90-99 | pel Gans ¢, — [oemge <1] Looege <1 | Zi \ Gath = tue: al eS Nets to Oscae ¢y 5 -tor Yosaqge <3 6 \ i 4 oo CS ar eres] | Fe Eleni _for_rumbu Ly siae Aince lowe Trench gaintr clue bh He difference we Ti} fege thom {go we will te tumor cil —didnlh pure aad eveddully prs whole tree + Noolet= celting 26 dloun'}-furn off prvining if dé Herence become negohire_il ary way Bill pane _ ln oth band - nel dad whol tes pnd do 1th prever [-nverfting oinng dele — - __ [ | Dooqe ie] oe ee eee (jsarf Eee “fo coludeh oulpal = (lum of Rerdoat Numba of Residual +> _ “olp4e dabdelepnk = mies eR| / | ; | Cr) | I Crt as =o mean w/o requlanzeh'on = —l0-> Tat Se : | if b> | uth Tegulon abior! factes = ee [I Ao Phun it will reduce thy ammount thal tae, | individual observ hon adds to ovecall_predichimn thus (lombda) reqularize bon poromeler will reduce Has prediich’on senshi to this 1nd in'dural observoh'on est - D 2+o | oul pudo| lect Tuber pee ofp of wale is tam o{ {eof -outpul-ot leof @ . -4¢ [Aiace we hove Tusi[d) new tree lnk Comme kelned dichm Like qroditn}: booshing we have tre earings rote | Y hece_aleo_, default Yolue 1'0 2. esl ti tome | _ 2. inshal leof + Te om ng Ao K DT. qT Lees meNow wert! busld enathyy hee hosed or ned yuidual Gnd moke new pyedickon "Peal con = ive Us Wer Jmaller ratdua) ond then Sqain es bre ce hosed on newel tasducf nd we keep building trees unkil redual ore bupercor, — 6.6 we_reacl! motimucs pumber . i we art ale similoly ve 1coare - Gnd ade +o d cteimy — how fo fo split +redata - s - tod we fede tree by cal ite en 4 —_hebween Jn, voluer_ and yser abe Ine bree _ —tomplil po ¥ if punplee fies de oot pune (iain -y ao she ke : negehm then ‘ er w_« taltupele Sain fn om Requlerizetion. paremetey ard when 9a Lande: ee 7 dr >o i] yew! tn mon pruning h nin bin suore Gnd Jenaller oud ve lice foe Sim ilon the [emuts - ee Hine “Store « TL Lom of suidue [e 7 No of sutdual +) — udp value dum o Nes widua) oe a Numb 0 raidual —Unsupermied Machi ne le aan - Clustering : his basically Aine of unusupery riiecf learning me tho. An csupeeszel le coralty methed ic a method fo which we draus reference Aotorby lonctsh’ ‘eqg—vfinput_dola with ouf Lobe ee) apances -enctalig ib used a1 a proces fo find _ meant ingtus | sbeuchure replosolecy Yoderl yidg _ procusces gennok’ ve feofures On aeeneny inher eat no sel bai trample. - da t= iy beat af datding © pupuloh'en or - “delowink taf pumbey of groups suk frat sbebapoisl fe yom_group core mow _timilor 4u_other dala poink: in Ho anne group and asesinailar do data puinl- jn —___ofhie groups otk a side an a eal h haut —ofe vi ont on t + und L ieil on CLAN 40) - — ba fine ha hu voluefor te ceahpatnl oy Ceabo id by iberobive protest y. Assig nie dtdoke nuinl fo ip cloak hceabe ~ the dlatpaieb. whith ore 4 pth ofiaennlan Jc — _creolu a duster. ——_——— ——& How ~ How does the k-meoru ams Al gam tao work d sheptt= Seledd lect Hhe number of Ko fo cleeide the n ——clurbers — 4 —theps > _teleck random point on cenbatde — stipes: fA aitpacachclslanatet cs thie clases} Cenk id, ned clus ber -_ Sheps ta hele [tu al « ine ce om d__ploce G new Cenk i af Df tod clus - et Bayes He Had thes wohich mens teasnign seed - dal ecch _ ___3e Drawhack : Before ctor hi the p 10 (AL yon yuk poerfion the number of cl ute, we _How +0 cho rouse hes ett volue of K" ko ea dai” : Elbow Method : — this method uye the “coniteph: of wess vol Wess chonol: for within cluster gum p} square whi __ defines the folal vorishion within cluster «the Armulat: caleutote the value of wess Ire tee below . ( for acluster Wiss: Z Piklastons Mstonce (Pres) + Z orJelusbera oishance (Pr ca)” ——_- * 2 pi/chutens ditkance peetsytoer L f —————_ \ £ | __plobc | __jhe ip point of bend or a pork | broek _plof | ike an i 7 ar then thot poink toanidered ay bel vel ue. of K x SSS - whee - =z Pi/cutat littance (Pic): il Ve dum of Jquor of | distances between each ealepoin} ond. iL raked within cuba 1 ond Jame tor other tero-lerme | ; ee “| —| , A agdtadgtdieed + 4 aa i ete measure the olishance we cor We an ethod Cucled con meth - fF eft ghance 04 mooho lon dishance dulket Oe to find opkimal_velue oftlusters Are ‘elbow meted poiloe Ckectibes K-nacoas duskering ona given doloxt “Ar _— diffecen| k volues {-vang. from tele) — for eoch yolue of’ taleulote wees value uive between eolculcted woes ond Rereber of cluster—_D.@SCAN il_ctonds-fFor Density Based ¢ pabsal clustering = —o} tpplicakon ui, Nohe. . +—wheeever we wonk to _understlond shape ond thudur thee DB-Ce@n by usef a ‘ kemeans ond hreorchi'ca) cluskeriing both fas 19 —_ reekin cluster of orbitvany shapes they ay Bal i. densikar Thoh able won _tlasbern hated tn vomping — why we need pa ccan chatering - wees [ere OT ° . “ Gp ___ § o% oe a ‘ : ry ee . . 7 er . ft ~ mea sss - : — —I 1 = 2 The main idea behind oascan 13 thol point belong.to a cluster pr fic close do many pointr from thet Cuakey . there are two lk oromeley of Dascan. | ——1)—eps £): ole called ar epsilon the distance tral = _Specifa the Avighborhood - Twe point ore tonsidcred. ty he neighbors if the dubaace beboeen them ave les — tho» p« equal te ep, a YU Mite phe Menten dumby o| dolopoink fo dehine ae Roswd on dts dene, poamssnelen poral ore Neasifral (ole poirt p border or boundary point and Nore Or puflien?. _ — eeCore point oul fa J Nowe (ore _poinf 2- A_poin! tr core point if there ore alle art _ min_ph numbe of points (including fh point ihselb) in th Jurceanding area with rediw epsilon ce) —_Doundory point 2+ A poml is boundary point if) ro reocheble om core point ond thee ore less mein ptr ——_numbero} pt peinh within jk surrounding orea_ov irsimpler words the poinke which covered the (ore oe spots _ se ee — Nor e/outhey Z A powl ie onoullia if the rr nak tore point and pol teacheble prom any (ore poin. 7 - Pros on d tons of accan pane — Pros:- 1) Dow nut require to ipenty number o| cluster befouhomd . — - ) nce eobuel tp owl|i'e7 ond ble to debeck oudlicy. a “4) necfitmn ‘aan tai Ph pbibrar shop es duoler, ee 3) ike Com c be [n Jome cares _dehermining oppmarioke distonce » : neighborhood (epe) ic out easy cond if require dormain knowledge. Secluster volrda hon Jechn'g pug Son “clhow method Con onl Aone for kKMeons but ___bilhouelte score con he wed everywhere __silhoue lte cocthicierl “cilhouelte Coetfiien or sole ha 1 onthe ¢ ured tor caliuloke the goodnew of - _ glares fechatque for hel blesler didonce between - the cluster rind he ss tele thon distance Within the clutter thy value Jonge fom =| fo 4 Means cuilur-are well oport- From eoth other — ¢ corly dicking wis he eof, _ ee cluser ore indibter ent: Ar ist ton. say dickence O pelween clusky fine 7 con =] Meany clulu are oe aig i any silhouette score = _ Ura) _ 7 ft Max (a,b) —_ — wahetts —_ _ (a: = avg intra clusty distance tre ¢ the overage a —dithence behoeen each point aibhin a (luuly ba ory Anke clube listonce c Org isbante— - helween_aut Pre_cluctery-_ _—hiverchisa -cludering 2 Hierarchical clustering i = o qituesiesed math ine learning algori fam which, . Tr group the unlobelled dolovet into cluskey ola sali 2 Known as hieror chreal cluster onolyss 0» HA SHOR Is in-form of chee and nis tree shoped shuchy < Called Qs dendogram, ; : 8 1eulhof Kmean ond HA “mass look timilav bul they ~ both work ditkerertty As hee no reqw'ement of - a {aa toe number of cluster Os eta 8 Hem _techniqun has tue Oppraches. 4 Aqglomerohive:- ik ic bettomup epprch in which _ Lolgovithan stortr with tobing at\ the poinh* ay single Cluster and meeging them vab'l} onc eluito ts left — a. Ditsive. i ibir teveseo} ogglomerchve_olg othe “oa ibs _dopdown appodh, Tt Hot works 2} _atee de ~ Creole each dole poial as single cluster _ _ for An numbu | dlotapornl_a_ numha of clustou. _ alee a Toke two debopaite closes! datoporek or cluskes ond muge thun fe fpren one cluter Jo ree will — i he nt cluster. <= — ee hep g Again tote two clorat cluster ond onerge Thum tha cthey fo firm one chute, now thee Will be n-2 __ huskey ee “step 4 i Repeole skep 4 _unh'll_ one clushen (eft __ he £. once oth the elutes on combined] infoone Pig cluster, devdop olendogram do_divide the — — husben os pu fhe pmblen.— eect ee— Meow of diifemee hehoeen hoo clusters s~ —dinee we have jeen clorert diitance— bchveen duoy — elute ts vey cuval for Hck thue ore Venuuy _ 4r_caleut ale theclistonce bchoeen swe clutter, ere woys deride fhe rule for clutering, fue prcarurer Ore Carlet a Linkage madras some hs nupudar [inkaqe smethacl are giver belo. ——leliegle Linkoge!- it tr shorted dntonce between pre cloiys en 2 _tymplete. linkege t= ib ts forthel distance behocen tet etc point of two tlusbure ibis one of tre ___ilt [inka ey tran Jie sil form tighter duit al lisk oc a et ' J 9) _Avtroge linkoge:= it is linkage methad, Trobe Am vag € tf £ _____distonce hehoeen toch parr of clolopoin bt added. pend the divided by folal number o| datepuial- to celedal ary olubance betta cea two wifes ib ty — elm one of the m0. sl_popular Linkege method \— tr — } dae4) Ceohoid hinkager— ibis Linkage method rn which the —distamce Lehwten the cenboid obthe clusler is —colculehed - ee _7 A priori -Algeri Puan ie fre queed hemsef b ——gerecol atsocioken yule ondul iL cletgned to dork ——orn the detohaser thal tonal» transochon «with fhe help ob there anperchen ule ik determine pow — ae —thoegly oc weak tue or more obfectl ore tonneched th ih theabtve prbcen for finding The frequent — —ibumsd foes lorge cbobaset - Shy 0 Api: _— ee oe a ___skpt: dDedomine the support of ‘bemet in the —remockenal debebore end select the minimum ——Juppad_ond_tonftdence a ——__shepa 2 fokcart the support in the dramtothon pity —kigha so pox! Volue than the minimum orteleched v Apo wort_-_Assodoken ule_M ining. thept) fird aul the ruler of thi subse thet hove highs — tunfidences Vol ue thar thy threshold oy minim Confidence vole = _ — _ tel uedershond with txomple - a Mik bag bred Bue 2 Milk Buiter egg Keebchup 3 Bread Gutter ketchup 4 Milk beeed bute a ee FS Breod buster Caukis _ C Milk bread batten Cooksey _ t Mlk (ook'a ee - g mip breed bub er — 4 bread bute 39 Courier ; ro milk bull sed toil pred alte© Yee objeckve trto we thir tronsochon da la ir to Find offinstin becisean producks thol which producl Jal baether often the suppor! level he a 3a°f+ ond confidence al soe -suppet + fre wee X29) % 35e* p equ - confidence + = (419) nt freq (XD op t _ 44 benseh feeg _ Frequect Like cet fr Fequey, Mk 4 — Milk — “breed to Bread ie : bulig No the to faq ooking x Kcelehup < = { tri rohiw min supp oolwy a a ae —_ettan bought fants _ : : toto! Ho: of trams oction fe slop + From thir ane theen ad “nib we we will ce [cul ole fre vequenuy for que item set = = a - a them af Frequenuy “feequel 2 Frequany nt becod pe iberntep a milk babe 3 milk bes of + 7 milk, woke 3 mit ey butt A _ breed, bulty 4 breed, butter breod, Couks'y 4 — breed, toh 4 butu, tooloiet.— 3. = based bn nin JUpportskep 3 Now From this tua | ‘Lem sed we will calle se Prequesy — of three ihemsefe ———vantquthim ares sills, Qread, butler; G tooltip ge -them seb Reqisemet- Fegueny _ milk, breed » burten trill, bread, Cooleter 7 breed sbultur cooky = Qe —britk, Butkec, dookyes a = Tored on Suppint thruh oid ce a : milk breed, butto _ voli ia with sup Now tk ime +o Half ausoeichion paler . _ + Jubset creakon = lesnseet a-ilum sep +t ey Milk, breed) breed) utter at ay subse} ore ___[Mitk) , Cbreal), (bust) thread), il butte), Chreod ‘butte ) . _ “How to form aasocok mn ful ules: for Way fon- empky subsets sok tthe associ sciahion rules Vy S— ¢l- +) if sapped W) /supnet (s)> = in \ - Curidence . — _fivex milk 3 (bred, buln) _ “suppor (mi tk, bread, bu Her) so ak. inofase Support Comite) _