0% found this document useful (0 votes)
52 views39 pages

ML Theory Notes Full

ml notes complete
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
52 views39 pages

ML Theory Notes Full

ml notes complete
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 39
; _ predicted VoF d Linear Reoyvesston! Fen pete Volos. eR Concept? haar Reg retilon ie elebistical metre, ysed to weeded the vel atlons hip b/w a dependert \avwable and ene (ov) rove (dependent Variables Lh asus Oo liner yelabiousbyp b/w fhe vengbte aad ales fo find the bett- Pitty Whe fet pupleize the bts boner blur the’ PV and PV. y pret Velee ' pest EMne Ao a seo A \ J% sy jatercgt >x oe! eee Foymula’ PY ex ve | ys depended ms slope x ~ fadependeud co: fatercep * Slope £ vatercepl ave very hnportent £ ee values Cheull fe accurate: + hey will decide the ft lines , [ne 2h y ye deprrdnt jatercept = Slepe Fi x TJuvolus 4 indepertet lependest Variable: Pu Xp PE wove than one tne oy Bet Ect Babe ~~ dep endo patecept Fe Pra ; Ge fr by | u Grew ae 1. Mean 2+ Reet — cle p= noet ©& Be veto. prod ae pretncs + Regen OO Kquard Erev (ws 2 pre Steed Ere ees ftselte Evev CARO Yo Re Squared Ce) - Ss ee Pitre = RSE = Ty (5-3. a fe SSR — Sum of Cenc Regrerion easy “3T —s Sum of Guava “Tobe p. £ Crg\h. a | : 26-5)" nee st 2 Las) | “OS [a | I Ie | Peean Phsolute Percentaye Ever. > Ordives Leni Save Cou), | , B= GcTyyrl ‘ : | aeety Neen) —Cbaceyy TM pie Ge) oe nex “)- Ce v4 . | Qayte ster “ihe i at x @ Dew (2) i ae complete ij ~ Wa etay uf Gretiet Deen a TE ts an algawither thot sfieds the best fit line fey given taining Acdetet ina cnrelder poof th evettous Applicettons: = Seerol Sclences ‘predict quan seeve, Matome level. 2 (vance if Pee ditieg sfeck press fa ‘a ntege. . - Foy be under bud - . - Can andl. large clo ter et ffty- —' Eosy f doplerel -Outlios ca» ~ Mey wt caphee. couple yeletion ships L of fot “the model Performance’ (od variables Yo heristic Romesrtant aise Rees hogis tie Regreetion Se at tee tistical meteor uved fe Prectiet the pre bability ef & bhoory ovcltome . (fx yerlne . ty ~— Tt ts Qo Claes i fteation eljorithun - -‘¢ Shaped - \ t ~ Oubtcome so (ev) } sf) + ys dependent me I nelependent | K we ghee a fhrerheld volue aud Wwe pan classify fo which clore they blog Diff ble Linea £ hag te binear pare alee 0 ‘fev input” \clata ee oa > ) Le | -Oufpte ~ ofp si cateprieal of; vous | veel! ao Ce wed | Pod Sp moid Punction/Curve Y > Linear Rezverrton yes: _ Binary Logistic Reversion: Prrdlietey wohutiy a crusher coil bey oO _ peduee or ret -ordered . % Mutter of bogiste Pepversions Ses . Precliotteg whether a petcen will vete te forties: 2 Pou-erdered - = ordinal Legit Reyverten! =). Customer cate fretoy bored on tKerr feedback. ~ ovdered- Bpplicotion: ~ Pedal bogresic — Email spom filter9 - ae ; je oletareb an handle long vr 5 Neate pix ~ Provides probabilitics , puhich can be usefl Je cision wank lg = Dulliey can ~ appt tea” wiotele \ Peformance T Mey vk capphne conpler wclictiorebips Noo Po | ee J, J ERR (einem wot) NS Deelstoy, ir Tree Meriter: fh dlecition free 1h a supervise f thet can can pe used for Got Cfnecificction PIM Cased i'| ‘ i learning oo at woke , : Ui askieg & Sertes of- que ters, on answer, we make next quested» epost node i pega [=] | oem par lon ice wen question ov ae elect rew Na be 22 bree - i pedes Repo poodle yo A ~F Root weds! Peprest pode, vepreren’ potter rede, vpraen® corelesioa [prectic thon. caf pede! heaf Meds pets fon ore yok * £ arctle node [aiblad queacton. Branch ! —— gael proces of 1 ofividiog node jute 2 Col) pubined es. rot fists Tah ecec setae] = Vewmowing — Uwnecesion Creuchet SAM neder font 160 bee fo yeduce ever fitting Fepformation Gain! \ rrenrime wsecl fo cletemine the elena 6 : , pesporfawce sf 4 Seateve in splitting o rede. Cobeuld be high ) . © Giei_ Sader cDadexs ft. promis f opunty sed fe spt te alata info Aérrezenons groups Cibd be fers) - © Fotos: Uncerbainity huoth be. lore pb we wied Cel Seno fetn TR dates Correct Ce) not eae ee ee et ele ee a é HS yewlves around bps aX £ Enter] : Z phy, CA) —avevelee avrend m uw ncer teinity » Dan formation Goin! preted — weighted ong Bicy Cenid Agpleahons: aC —HealthCane. | — Mevbefins . ‘ Pacem Disedianteyey ~ Fae of Use Aye z 7 espe y - Oveifittig - = Black Bea atere - clake “typ 4 “Ay ae Feet 1 Goritinn Cree) Gt) crale rhailil i TEI Cael for Batt O aggefretin f Retain, Ho— ce. | Clewifay to Decision. Free Atgerithm - Beloys fo Zngeuble farguaye . rT KIA combiner raulliple meclely fo lwrprae Te acesvaty of" Preclic Hany. Fepenble metbed (Pov ate) ay xotad) Begging Boosting SbeeKlog | { \ | Bey Te creates a diffe ay cake ps \ sample painig deta opt Yelecemat £ The fir | oudpet Ve bored am (Paybwily yveihg. | Bx! Laver Fovet AAG. i Beagics| Ay fi] 3 @ ai > a) OTOL : Trey ae os) : da- Vet Ovbrone Bootsbap prodel- gemipl -— pee goutig, TE coring weak fexenng take boy lewnnd — by cr ebg sequerBial mecled- Ex! pen a pardons Fovert Ae Peciinteel * Pedsin Wee , Dede Wee Result » Renfree co ; dS, rafority voting, _ Appl cabien | — theatheeve ) = Tamoge: processtag. frecarg > Veg V7 UPR ss enlthel. Adu g DS Ade over biry > URES , = z = Conpitet 2 Quick b Slee fe cio po ; geil bela - DE nts ox be elakesel- ' , y KRM C Neanery Meighbar) Pane he) . K = User cle fieeet Ny pet params eter — Rela Ceti Preeti £ Repreuton. Te fivl Kk ualues VO = & 3 “Thun brate > Sampler are Vezetead fo (dentity the (UM. DKON works well pov coyaller daferet~ > kK Fu Sercle fined hy perpaameby ” teat deterring Low wary neighbour Le consider . The acturog ts saturnted’ Phen it tt gptiand, Ke volue d rf vl pet oF . — elec Prectten. fry of the Begav7 Kur” 4 Sicple ster’ © chee The Solace of &- \ = Cakentate Tre afiste Glo new f cobsbirg lake pent. = find pe elemet late points — fr clesifietion, vote Fr moety Te ; Eveledion dts bance foresela Pic fderrteges! — Trail & errer wmettol = —Pime complvity fF mere for lager Labor et. ~ Tf K volwe ts covery, whele prediction os woverg . KON 3 no mheinieg pretrod (Lazy Learner) 5 Manhattan fercte pe lye] oc! lan eee * CRD TCIM)” u ) or onl calculate fr beter te cheore Fas afd » Nalve Botley C loess fier: 7 vi if STE te a Supervised (eornieg | Clossification toskr - Agoriten re poten, DBered on Bayes Plreoven Ir probly - Swale’ means features Ave fncleporelert > Fark Clorsifiertin — it feo vil seeler cf afacel : Formula! I i aaa PO) woken, ; \\i fre Beet B > Evidence rWa) = Pestoror Phobabi lity . PAS = Prey pap ~ Likicliheed - (6) = Feideee- (Like hood x Prior Postevoy = = Feitence - yee m . Lipend plumy ee Rat tC Pm Pla) ~ Bide Types of Nelee eyes lb neon ' Gavetion ofielereninl Bernoutlfs | | moran debtipe: - Disevete Court, . - | - Contivews ctatedgp — Bloor dakly pe a Batt height wel ht : Al — Word Conk. — word precvedf 5 obra < key Chaveters bes ; RS eficcune te fenters follos 2 normell dicbrbution. = Deals wit® CourtBered fate. Appleton : Texte Cletsifieston - ~ Spon Filter, —Sectlmedk omalysis = Medico Dlegreny. ~ Ghether prediction. — Face Reugaidien. Sberg ts: : — Shep leity ‘ 7 Becee entene tncleperdence . Spay pea — Good for “Tent Dok — Heveller bail al imenonk lala, Lilet tables: = Acsumptton of Feotuwe Daclper nce . ~ Dale Seaveity- yi SRP Vee bon Machine Coun) . i nae Bete C Cassy Picistion = E Requanion £ Ouclltey Detwca,, oil COP le Steg l-- DE can ge Compe tos fer Deep (exemty — “Reenel trek” fH | ~ Maily deals };— Ay. Is ase & clescify a ee PUI mick tr co lex. ~t Objective: Ty fo fred = Plane tat hop rae matzty Ne; The par distance Bles ‘pee dots pole of Lott | oe eens ‘Ave bypsiplenre Which ox ¢ lose b hy pep lane. morgias Perpercicler alts bance fren ey) to fhe newest cfetapohat of ciftor Olosg vd margin: (Jo retsclosiPenthn When dee wi'tK bed igh: A deel } Qeppet Verbs: dd cerns Bounce, ; fat Selassifreation when tong ysith te Sft magm: Mise Hime Complenity . Non-Linear SUM ~ Et pe porcine Kernel fone Fn. : (P22 polywonrial, vatiak bess fine): skermel_ticf trick “Ucel Bb Penfore 1h jupat space fo a | - A yeolly cbernt fraser fe dete. le; -DtosD = pt put yieolize fro clatr inte hagbier cbimecrion aud gum Bie 1% Sob : —Pelyromied Kernel: kL = C0 xT x pr)4 KH x Extaw aeta point — Yje geabig fbr Hae te blog: - dd te degree 6f Ste Prlyrominl Se Bate Kee exp © ll xx! Ir) Vv 7 “hens 4 stmilavily py tue dabr polaty jin jafinite ve f thee app teach a lewificnllen by waojoey Kg xt ave dala point Pit a pammley Phat conbole Sapred” 9 P cg fo Reract. aud | I k Gest) 2 tank CL xT eer) a Secllng , Lclr. Yv= boy. ag “stzetod kermel tre siedilanty bl Avo | alata peinb ~ ic (ora fon <1 te J Fb prbeed te used te selecl The eptiml vole of K ie woef Cleates CoD Cenbeordy. _ a — hey DE the cata foe finslitg the contrid. _ Pormabion af eleeten Bevedl on certain g lenilawi Ever — kemeas Clistery ele Kou ay Fetal cee’) Bin ~ act) mae faci), BiG cies EG. ebhed: oe Z Wess (p= Zz yn ZL ye cles. ay Me Cukor Sam of He Spuary t| inane Co mputetionally Expenie gol | pattem Reaytation g mb L—»> By Btskep - ph sees Vv Dewily Based. ee Dirodeort 9 Fixed cecef elutes): = Overfittirg. — cakes time i fe bain lager cfutwet . bail HMesouelanl Clorter'eg: - Unsupere reel PAL tebniy nn. — Used fo grep Statler daba Poth pule'e lahey, Tt builds a bievaredy of clertey whith ca~ be vepra~bect evga eee tne shvelee Cotbed | dendye | tendre peace ] hee Setiney of Ml mogy £ mplt- Tee mettedy —Ayz mative pre thal. ~ Divibive petted DAtgemente Kags) — Bite te bye meted: WT SIRE ite ceel clactr prado thy gg eloré labs I A Pteritheg “TY fee closest Clestny Style hota, ae hy. BL fBDE uHbl a i 2 Single Iakge — prin Complete» — wax py 8 oy lowe pot ae © a= —3 6 Se £2 ,-, oe Gc ovgtel Vitek, ATE fren stern phuine mates — (28H) —Tep te peter appeel He the al ter dete bers Ih sigk cltey g pa ftereting, we Pl Bx tat - ft fee cod, un gel MW clade Qiinte nehet Popitcottes; — Riel foomeFie. Deen at Cosi fehin - Drage preety: gapecetted ML Lecbniqne _ This fetheique ts used im pal Jo bepoue te overall pew forenans® by Combining fia prctittoa from prultiple poo! kr. oping fe hope Becurney, yeduce yeurimoce ft biet = grspived by toe wthdere of fhe ceed" concept auly we Evsenble fearty? —gepot Me"T efter perPrmne Loom Aadiodeal pastels. _ Robestrer) pee sloble fo yieblle prvither — Reduced oun fittlrg Cor Ow pudel gone One Leer a Type “Voting Creation) - Bost shap Aysgete CPap'D - Boost —feerdane Forest” - Stacked Gersnlineliie Lilet ») DVekira (trmegieg) = DEANS an ewenble ML af: tat Ineole mae Predic Hon a Hot & tHe avg Cre yeast) 0” Be say Co losific ation) of multiple ML Model all y Fh Cleseifiention HH tb pra fority PeP hey el ~2 Mi Te Repremsn (0 faber fg voting, | [owt tether Data peteot —— | —~ | pee | Pred -/ ) [vo ty cenit) 4 =.= L : | h i - aaa i | WW oo I] > Scape: i Et crate a Ah Beimtes oD fom Sample pewrley faba wil’ veplacenewt dh tle Fal 6ulputat pee mee safety Vo beg : Ee Radon Foret 2G nefits ~ Reduce Vawiance hk cwverfil tieg + Fubaret pectel bebe lity Favela om Tovet s wd is ured fr Beth Clout Phaotion AR “7° IH0n, _ Siniler & Pectin Tree Ag: — Bey & Exsentle loony ' by , Pechivartee | Tree-e Trec-3 - iS L a Kesult 1 Kerult fenton pS | Prop evity Vat y- CL ek Fe pK cah'n: — Helth& Gr- — proge Prceoy free orev ed Foyt Use — Feet po Trin ~ Uf wovks ox lage eae ao] } | —Boostlyy poubinen Tle weafe leanne be Sore, | hemes hy cventieg Sepennntl'al podel- | Ex! ADA Roost Send | Comret GY) Dicoyanee™ | TE a Sop entlal meder. = Reduce bth pike A varoweg. _ Lyeprove aceuroed on corplee datorebs PPh _ pdeBeast = Ga diel Boosh'y Raggi we Hews LheA Cage tee - on | Checked Grocers ?P feneftr Spekid ( & Grovral of Blomdig: ee Cteck ta a _Tyatn na tiple peve vordeG — se therr frtitior en (a pu Ferm pata nwnndel» — Melerrodel keay Tee bot war 0 Comblue fore porodibars “on pase model on pte baloly 2 he a seperate voltae» get fo thats the prelate = Corbina pretictiow yleg be pocket Pa ZRUIO pegver to. / fred 7 Fixed vs Telned ale Fuses | . i 1 Tlexibiteey Jaw WW } y | Compteniy § t nple- Cup nak of pall tel mang Oveflity low aie ~ | Se = Cure - of Dierses nob oe s ne | — tapers when dedeseh have to wns Pankevey. © Maker Garda for wadelr fo cook Sfbooly, tne soup Ce. woe. Tener Coup leuity wonker oka snc f coupler matele Hl Why wove dhrewlow Cause prrbleny in le, is Dek 9 foo cprad cur | Pattery beere havdate fin ¢, Move hevelvat falery ead fo Confeelon, — Crug ulation, fake forger: Peblens ) Dake Spacity — Ung le lad Patter) | 2) Baclenct fentaver ~ Cone overGit) . Wy Db Ane bene Mea play — (FPN will Cbg zle ) y JPreresred Orprtrtlrn = (Talmiy podl beeme phi bi Kaos ee CP Aiige Fae bees “fag Dincrslons Ciel festory [he 09 war © phigh Obnvealea@ Pore fais shrackd ~ filling foo woany fonte How te Hawdle te Cuwe Fen - Dirreas fovraldly - Kepularivabion -(Adhe op oy deci t hel of Dlmcerlend'y dave Selecblon —Cret ft vadabler. Bx Reddbiey bote prices Reduction = (hedachey (wo fer & Io bepany, cvlty for wily be may Pobre) Lonren teebmiger: pasa, Rigo Regzritty An (gust Feature Selectie Tee —_—_— - — . Lohy Feabuce Selection = Tepe mdeh aeerey - Reduce Qverfiterz: = Speed op model Prog “Type of Fenton Sebedto oeiite -eloed _ Reduce hionersivolly es Sk by cicsiratig jnvelevnut- forte £ Jafrerence. ~ Couvelabon Cefficrt Wreppa Melted Lhe'~ Square Fat Fe bedded Mettah— Pebid 1nfaveat Vostanse Flore Forwad Selection ae aan Bockuoid Selection, LoernCes Regroro- REE ( Recwsion Ledoe Piirncbs Rie > Dees ion recy. ». Dire meted, i|| ~~ Seleet feabuvy bated on thelv indhidenl velationdnip with te pared vata ble. By Fray D Corvelation Conf fcinit' eee = fBv Conbinsas fete. Cou Oo eK Ny, - ey ot Pees | —Meetics Viney yelu eas bp reco nabhle. | DLMSquane Feet: |

—} witht Reger ade Geta on = Covet) G eel Fe) 2) Pecttlon “Rees fe Frcwitor Sele tint Selet aati besedon baw well Thy af le be eee Gain. Enbepy (prunnt) — boeigtted Ary ebony cceild fesse Us Reape Lotro Riles Fenty — aD be Fontes Yu, seme toeffils | 8 all coeffi Slee pe cot te new | cre sheen bel vewmate iM ba medal Effect on Coefhet — “ t " - ee Pech ~ Paci Corpenest _fralyst -Unsupo tied Boer, ob es limerstovelity Pehatuction ¢ th reeled. ps ct te et tw SO—UG Too bnige te sol tata conrecbler detest vobile si poosinboing shyni front potter § br: — Suvalled dobre ae cnsior te expe d Viele £ feng pals Analyzeng AAR poh ywahoeosy £ fees ter for rl os fiiturpto. of POA? aS = Linea’ = Varlanee Toqportance — Uncewebcled Fanboy Prem Content pata ) He PA wovk SEOM SY cruchird rented — Gtaldaidlire the naiuge of Ceufinuny (nibal veto G, elltc-vr)ens SFeicovd Aenebla ZW STP2? Coveniane pate Zh be Stop ° Wotls Compulrtion — TO unde bend bows the yaviobly of The baprit dot 54 ane vay og fren fe prcen wil vepect t2 ofse [Covbur Grr) Gree EXCIS Corl 99) Cory es Cov 2,2) Covlery) Gvtzz5 Con vretetet) af +e fre: , cof Lhe boo iaubhle Nee) 1 breting Th ve teen fine Te Wher the Oley (6 / Fav ew cGy ELF Compete te ecgen vere £ ee gy _ By veuk thy, highest B loot, you gor Prteiiral cangpmnte ts ov of seguree- SteP-v! Crente a Fentwe Leder Stor: Revast fr Poe peg be FoAza prunl Puesebs Featove Veep te Scharber! On inl abetet LOA - Line Diverininert Arkyst's- ee Dircuclonell'G Reduction tech nigser— supeowred Clout frat preblor — ge ta = Conmrerly Used b pret Cle pow toy 1 wed fer lows in bf invention uke dtprcantha $2orr x Two evlteva ~ prmdeise Do Lttewe Gls premy mens of boo Cor a Mamtonize fae peroletion gottela cash Lag. lA - —