Machine learni J | a. == = f Crainy q Supanied ML ; Unsupervised mL. ear ‘ aaa eee fo = clowificahion clutkatng Apri — Recommdels. : Logi th’c + kmncanp kA soascan | Conterd Meno e hereby | User _ilen bord bareo) Dimeabi c v pea tene [CA Sv Fuh ae = Encoding calboost , adabocs | = Supewied ML —> fae short with Reqreatton =p [incor _ lincar Regraion T J Muth'ple pelyncmi a) Regresiva | incor Regression Regraiton ToL BR) ?> whe for one fertin then i rece Regrusion (SLR coph | packeqe 2LPA 3 Lea ace a.Basics of Iinter Regression z a Equehion of chargh line TONIC a fied value of om le r Pr (Co, 4) — > x, vy Vig (BR0p i) arma Mg 2 Seine slope ee Mae |= Grier Xr %y Baa 4 —— a iy nei Some pcan ai He een Value of % hou) mere eee laterceph ae. at Py (2, Ye) —y (yg) 2 mt +O - 4 = oa te | Bagie Cia 6 ! i ctheesabove line eq” is funchon trot telok« 4 ord for given valuc of x we Can ind otrexand ihe Value of. 7 whal if ie Hod 01 cance Haste chute indeacdcal- eee then ih ie called mattiple lineay regreasion ois Te (3p)pe noe Cc sple linecu reqraston 2. _lineas regresy'or fie wnireon ae ‘equ, {) simple to to aes ? Ee wellonthe dola with linces teloh'onship Fee paca 2 1) Nut suiloble for clote heving non ( tclokonshp- vader fitting (usue. Rriveto yu Lie emeh cal Vacl ea. ¢ i cobled bes} &t fine aoe? thir ae lee patho, ct Wind — az bel fitted Line (model) A for_lincas reg recsian th seins (Comes in mace _— : — | firdiih eine a fous s PUo-e cereus) ECB) (22-14) fe caseuy] - Lec Vruo eee aeaaha marie Maou a a f { teos0n fo toke Squar- jo thol posthve and pagehie} veluc_moy net cancel each othe Doig (oi vohte ——» High Accuracy é high [ou value ——> low: Accuracy ae | i re l adil nizohon. fee. fy mo cd ut giv inim um tunchion. Be tlilals minimum! io er. deacen! 1D opAhmrs«_pth wil | step down onkil la a = )} qlobal 1 minimum hy ibecohve (re ces < — a 4 . ey ry ! postomel « = neem bal tothe) Mavi mum Ps Cm36s) 9 / la machine leaning a fini « [em iow) LD m “| 2 w- Ld reg. \ 6, shev st L De a WERUE db Lin slope pints Ne w = werght : Ic - Mieccen b= bray. L = learnring rote : Meno gatude of chonge te yu want in pocomcter dhiveg theroh’on mihallomievokve o£ fos. abcient oimhon e hy evohve of [oi funch'om wor} ¢ L ibe! i Ueis funcheon) , 2 Hae. = ™ Br c= Phi Wings 3c. ( gi eS met. loss funthon) . 2 (yi =, za oe Baepicn/\ t=)) hs indepen deal Vartable _Y = dependert Van‘able +t sz | W- Weigh (o [b= Bias ye a ee rr col Shain Lose Funchon art | be b- wx dbs : learning rales - ib istuning porometer (non Ophimizc an olgonthnn that determine ghep size ot eock tferahin Lohile moving ctword parnimum of loess Funckon | eee (yh Seer | ws et oe qin loes findson wk 6 —______ | predichion & doetooe MLR eae) Ae Gry VG, Kt nxn +h. leuncd Reg ec 2 Ati ta te bagtat = tanya th + & +-- t predicled value ie Actual roy | Value . --——— r : em Reareciion ) - - HY¥ Seater plot + Vv 2 ro cg lincow VW Non-liacav(x) Nok ole (feokure ANetocgc) Mut Beollincon by: howd by inden on ik mean there at ov should nok if i gia eR Moece Foran geod . je a “8 suleshat )| 1 ae ; | sta chek tbe rei 2 1) vie (Vorvance InAot'on foctoc) ele eo res j bic qrourdl thin feature don't hove theissue Lomd i} ib ts ¢ or more than Ahot thie that E _pecbicular feoturs hove moulh'collincrily jasue and [need fo remove j E a} anuther method teh fie d_out Correlohon behreen al | th : feotura C Holmap ) ; _ CO Date: / / \ | \A Page i 1 Otmals Renduoleete | i sous error C Be neniateaD) _plot_or _groph it should frtlow standard nono) I dice lah oine means masimum error should Ground mran =0 how +o check rile kernel density Pmalneulikelle ) Clrompl, herbie) Xe apo | (therokcal Quonhi ) E Homoscedasti ily. j______ ame __¢catHere ik ae, fered pecleg Jeatterect if Te tenet vt ee(desiroble ) (Nel déeivoble] heck:- Seatta plot (-y wed raiduel ) : poukve Auto cordokon Nege hve frstocorrdlekon— ¢ dex rabte (Nor desire ic pil: plot (raiduel ) fois predichon Tofecoce High wv we 4 Mediu Pee i | low vw a Uva Yv Wa |es oak ae lal coe an ine whol tf date tu 'n ne I such (cenan'o wo exkact [ Cra the WAL a ot of inp) Voreble iin Bod lo 2 |x Hii wo we trol oad dale for ithe cuba A rons oats ie lnonlitnea relahionchi le Fete! 2 beta. eich reer ee i a J Ape fot Bix+ Baw Py digree ae So Pothict pone ps de ge ey ‘ i ik Now how we will tnow the perfect value fo the: r ll fos ‘not Pre OY eis lie! E ! { dy Ai'val_ alte + a a | Hog teat Lon 00 Variable giva wrt ini vL¢ Pdgon thom numer teak the errors orn E huled with Zere mean and tonsant variance and CA [thal thee te no reudkcallincan by Chaigh carveloh'on ) among Jasin preelert vowable. sd Lite geacohred | [eae rquare inter’ Lhe Tee era phon ore nok me a | a lel urdustand arty problem ae aloe anes ie ee ive lel ssa ge ealliaall t2] U6 | 16. Ate ins weustll coltatale Hh. equekon for thr bak 6 line — : tusherr ad Pee pu tal pe ee acti as pevtsble hy e pales gaia ce tae eee ee | Des eee Ce les \ Sits 2P) SNe Cie 2 nee. Ab bo KY Hzy-g =X = Ion qi 6-4 20-| 4 _ 4. q-o a (6 na _ ele 56 a (numb «of deko esl) mv b eae ic ee Go Bg] cre alas Ao mach i eee me”“8-2-4 1yr8cH (29) ae es oe ee } Tinea. equahion weshouwtd plug volue . 2 smth. Spee OE yz ew HIX — 6-83. a - 2-4l (%)-0-83 Eaeeeaig poe = acut (S$) -b-e3 ete wee Beeps = te-4 | 16. 9 ye: depedec! Yanobli a Piindenedat vetlel ort shahmode|-apias so H#ieoding dolo from ccv elt = pacreod-esv (than csv") He dehining fhe variobles dh L te! ].dofich O df 'y'] plicl)(Gre st adding -th tont font term xt a eee) a pelo ming requaston 0d. fting coodel — —— Faull ~ ten. OLS ai kt o— — 7. | 1 4) eet if poe} Ju | psal: i ea ie an —— 1°) et «6 07 ee a pal easal saan i = alcc ot C ») i VT Wal a ee hiatliaconhy occurs >[ cack aguarc ane Bien eVect ance beta) rge This besull tn hee Pees bapes of req ere Z {elguce na | Ridge egress Con (14_regutanz2bon) ee aoe eee tegulan'zah'en)_ os 5) Clase Net tambo of Ridge andl Lasso) = 1 Ridg« Regeusioo zea} se mg Beading meen your train ace 4 de high buf ae = a a " 4 jdbecieal of ret dikor change pte te ceckuace ouecdr thing oncems—cto_tecluts Jlppe— Cig lyin HDs ! ee e ae dalapainl Hur theal By me ges ic le: Lins sohich pan x raph Seough hala poisk ie) ee rf: ace kk ae 2 ott a eeaT | lee ane a Nod we Houece Convey our model ty choose Lp ani Date: / / Page A = Ls Edt Yio aut some eel ge vere eid odd pene Uy before We wr tt ad 2 HIS ea al ter | rn nei ae = hyrepotarcte, {alto / See : ca Now we will co | fox sor ae With _ege @ Since tine ts possi'n (ane aeies\eee through both point’. (6:32 2-1) > lo-g)~ | pecbecly trols why An oni . febdern (aril exerd ea (ROL Nie Ea (MIeW)eets Go gq) i Evora ine =] 2-03 } os = On Bee Sec ti eA [ne deal cm alt “iain, ae eee hy wee Called 12 ince ipower have mote than or inpu is perelly boot % (mittma?) ond for 3 ‘® (mtm,* me) in ee ft. bm we ‘ ri aha. ——| i e 4 <9 @ e565 io 4 x | bs Sty, pee I - [loo fe just enethix van ky oa ee : SSS jl oP te ot S ea Se Alin CK ——$8. = ome + Alida fwelgfealde- 4] wel In ridge regression for Gry voluc eo} A thee wee alay, som voluc fr necnomee of Ln pup feature ) In lone 1 yuu conkoeusly iacrean volux of A all tesba'n | pain you will xew yolue for some a} cwocttuc! which _ Lan not rospartenb go hea we unknowing doing fear Sclechta omd if 3 odvarl og of Lasso E +10 whee You ot wo Ag On. ) Pee nal temp_wi shautd archer ,HN 204 xy X Py \ 2 18 Ned cor -0-0 hen ihe cosbhiurmt are a fcched mace | Be siry or you We ind: oy Meera: ee walt Pu ia i Jeere are a oolusllYy geeerd Ware c peelize hn ore 6 eres, Hrapsdly , roe ite coubus for tlethng_nphinim —— e dla prope feohire delech'on: — ae fl Vom ton ¢ zs hie Gi aH Hing decreores (1) fo tinctease ta bros ary Hrgh Vonan « babes Petiavader filhay _overh Hily Pi Wey cm — , ora Cae Higkbia ig ie [Bite -Vortance Tradeott: al | in importart mires Variance) whea i} Comes -to uccurocy [in omy coe AA goe'the) J thee 1 trodioft behaeen model's obilihy to | minimize bray ondivorionce thick jy redirect +s &) heal snluhon for eo chee lee ebaqulatece Constor ‘ vaderstondng xf these error: ree in Ta the wedlebing and vodef; ‘Hind oxa edolont while écheining 0] 9 rdithon. fo Uedethond prediction eirorearab ri nth wee oe ahah vonana Aale foot Uke follows— | incall) __* Blas -Var'anu tradeoff - i} the al gv ton we Simple (hypolrars with Linco eg*) ee troy he an high bias end [ow von'ance ond thus tf iy _ Orrot preene. “| 1C eccor Ft foo corp lex (Chey pother T : : equckon | theo th m foc high Vanence ond food bras, rotths loHes Condsh'on pe ney) goer aill Nol perform well. Hee jp 10 mith ft Ry “ him he fp 6 ae tondithim Known as frodeet! oy bits —Veronce este : | | a with high degrec a —— A) € Feet v£ Rig plein Moe on (oti Siphon | Lows Fun chon = i meojyulds ace) for an ech bed volue fomily if: volue ; +jf we ore troinikg binv dls Phe inks paiodcliindeey) OF, 6.0 dp tnow which model perform. beHten and whi, potomsbus ore beHer Jou Aunchon jy useful » Clash Neli= Chi tbe) a Com! L 10 oh'on 4 I 3 hal sy Reg = 2 (iF) 4 Fle} + tcl) : | perodty which vitemspesed on por0"™gh (eta): = ee ee aqe of Lowa & &rdge Patina) 0aIoEET — farso Redge ey Lasso €=0 RIDGE Cfo0S = sey gdqe R soy Los. lded ix qring asso ty. eas eee equlanzabon Aiea i will wan! Neer hin 0 aaq tere Year gr vera VP) a inate Ride Panoliney ebou) Lang deem ann 0. Myo} OF o. ree Vrs plob on di fhruudt te renee fo overomi that we have technique Pealirisl Jmoothingr ea E . Saale hows rainfasl dota a: See en eae No nthe ja a a a he = —— a =| ‘y, Jig SAN aaate 5 As Fed N ae =. where Mi - Janu Mi Lehmey eo my - Morth, s {| __ +a ty E Yas eee Mae Ye a “t = | ae prinf dint Hay Wil ash as ede anil go thers Jolue writ remain ae ib ty ‘for window s'2< fintrhas and lard wo wil! [emain jeme . OY Age epee nunsher uf wiadow Ge nade act : nue Nin Ya Vy tp te te te jp PELE aay igh teh be oleh ; ‘ es tru 2 tY4 1 piney hi hp te past debe} ) 4 tute st TM iANnt Vs + | at Yat vy + ys- a 4 2 A@) - Mowng average. | Hest of ! 2 sy a | = pied ee : ‘ny of ter [treed, Jeasonalihy Nesic 1 ; nolihy A Nore subse ead From onida b have dela 2001 tn 2010 ’ eg we yas M)10 nes REG IW EH Bl 0 Hob Nie Pie MINIMA tere Chetty 3 3 De Gian tit tha tayee oo tie AOj". Tatetihe tog nit. «Te (o. wl BiaipeNeaNaoan-se- - - - * Ning? Yio ham S(am, $y + Feb =forecasting in cine serrdd Fist ace i) Cwhin you don} now whether bmx sentes howe trea Volt wreker'r model - (u'phe éxporkial | if is pechormed well wheether is both hrend ou well as seasone) Dascesgnals “fy then helt winter is het) rll ool Conor hed a eS Drow D, Dy Dx Digg Daag Dao temp aby Bah olde tig, —<—Vin, Todo (predic!) whol will be Ha ei oy tpl ce 2 Peon uihea dis Wass vthaka tiphnfoday Po : Q ctatihg, Av g ce Atte bisban t roo: Lah in_nok good encthod Since ecu cn cx a lot throught cor summa wi inten anc ther wou 1» Whalever tr todays temp EET same. ovo morro. 4 tday roo - 27° stday ro} = 24-": ae.gq dios ay liag (te0 ne Averoge af Ooy 4 ine 04 DANE be cambinahone ompon cd ay' (rage d Ly : = ys sth Le ie p 4, Aince ther it_po_pre tous Tecore) A Opp o Le mnok'eat Unde rlendliig pase el ae l a 44° ye Liag + Vy tiag ® cf tnet kha] ttre tin ! Ligg +1 ties & L a © ‘ — 1 ! ge a eat 4 a gt teen 4Essa een dicts kal en amll 1 2 eo £ Sa oe f jhe 1 er _ Ly ba lg, — Lees : Peal Las ats, 1ep Fisy aes mia ileal sa ae es Se a: [Ego ealiceys LGis 25. ice, pe aro 1.4 which ir not true.Real’ AY piel rtaale i rs |os tocar brs ry iw ad ts bud local avg ical — bree empecked fo change by Jame. dodo 'y ad eee, ( Liaq+Taq) Ea Hered is eect i Ba fi = Liga] ‘SS Ye ye are Sn : aoe i let aes ea ate Sineeerinee Sot ry: a dola i colculale Lorat Hed fee cl (tit tt Yt y.) af, (sb) Aaa eee im? "a le Gponh'al Q Double Expanh a . a oe, uel og| — 1 . eis (Beal o- a 4 ss pee al paceth)) +8 4 = ioe L659] 1 bs tae 2) 4 Ob pe LOD) | 1 1S 22395 34-4687 4°6494 a ]| vs i Wow we ui Elitebereetpulleeespenbiae a ie Ya Yh ws mi | | E Seep Se | 4 i ty — To Ta Te = (, = y Cutt) + 4 (i) i : eles en ae eae A (2)? yrs ps tee (lta (rs-1)- eters par oe. 2 | fy = E(bs ete | 2 (4) L | ed (retomwed (a2 (Lard dist = oete dt Io 2 aap BIE ae ne fbaata\ - Lew) (vst = hee) F = og +t O43 Ol Sb T asaad), (Gane ata) nee O-28lre + os (1043 Di24lie + 0+ Shere i By es Por eI Ae cuKy Jott Te OSta) Boal a(lov'8y 9.1 +h, { crs 2 heey) O-4i4os dt or (151444 Peete +o sage ot ig Ts Ty Tr lisfzenel tr Loprrase df ejay OFS oO: g1— ee US Deca) cabs cf = . « xpon haf Aven ca aif Abe esI f (Oates ss | | z { | ie es. Hotty winte mode| 2= which tonkair both tends. | Te eee A > 21 Ye: lo An Iie fucioge wt 4) In ee oul tren a4 well og Jenson tidy __ _ dim « SOI dole. — ry Ni a as og Y206 ft ! Ly Le La Lesata) vead! | ty Te Ts Thee a one dy 1S) Sao sey $20 | for 200 Months —> 16 ym ~ =a mensth for seosen | iby = pecrodie hens of jo muoths | let fink de ae double $e on bial pablen. _ hoy tthibeaching seasore iby Fram bm jerieds Nislz8) Dates aynlac sal) ety P- NavoSz09 [heoo 2 Y(biag 4 Taq) + y (ree- Sree 6- {Nate _Atace soos 2 rou= Sree ) i aoa a (tyaqg\+ Y, (Larse—Liag) PBs ay ; fae Hef Seno =i) 4Y (Se) > | Fro m(1) 4c _com ge! do ienow fo yeh - Lies we subshoct cipo From Yaoo fen | Now $30) we con do Yaou - Lace — re = Yi (S10 = In) + Ya, CYaee-Lyve) ae)T ms | er for Livo we need 6200 for s290 we need Lave q ey iS en iis At t Ms pe, tne eg O S:00 =Siep Bae Yoana) + Y, (inne te) =I (uagtTiaa tl, (Yove -Sies | = Vy (Trae) t+ Yn (bee baa) psonal i hen we cam | whol showd be hynerp aromeley app! Future ined Nob |-lork possible local Avy pif Nah wn ih oll future predichon | L if are Rs | [ard pacisble [veal Av Fem (nae. epee cs chan xd co = 4 : ri cad and local Average t trend | 2.dil Based feccastng a. eal then ow jie oil. oo ie a)ARM A hy) ARTA Trem | ze . Here dolo mut be se bonanp: ik secons dusbechabin of dala thautd be sanae ocrors the me Do Ne omaAohnAe pA ASS 3 @ stokoeorg) WZ (Note fakenen)_ , Aa | firture dole is only depen sidus foctor. Japtet