Module-3 Bastcs of teawning M — |The standard defn? Hoo G.-PEOPO Tom _1s_*hat_o_pansgt 2 Sf hetfonengee fox 3 ack Tyand ) wet expertonce FE, tet us conides the Concept Of learning as follows: el aed odera pute X -Toput space 4 ~ Output Space (yes/no) Let D-fnpue dataset (%), 41) (xa,%),.-. (tn, Un) tet $:%>Y maps the Soput space te _tRe outpue Space: The Objective of leasntng peogram 1s to ptck a Anchen gix>y to Approx! mate po =e | al the posstble formulae fowming hypothesis Space |@o short let H be the Set of AN foxmulae ifrom which the learning Algoe® th ahooses. LThe choice _fr_good when the by pothess ¢-rieplteal + for att Samples - | (Tdeal tasers “Taaintna bypotbarefiy) ‘Samples iCandidale . formulae + = I _Tt_can be obsemved that towoet fu clon _aae_dlepe Leanne 9 _ot [foioeneLeasing 3 Hypo theese Rf teonneees masa - absovt thre Let us Assume A problem psed&eting ao label tora _ ggven_ to put data. 7 Let 0 be the Poput dala set wtth both se and —ve_ Sai ; 4 be the output wrth elas O oF 4. 1 model ca the _stmple leavntag ne US 2NOS, Thte_can be put 7 4 Stngle Kix) = $23n za 2 bas idan Oe components 8 l—W, Wo,’ Wd Be the we represents the = clases Thee: the pexceplioo _model-i | eng vo! euuee r) fp Cboeepe leanings) as es | Soneept learning %s depmred ao gtven A Sebo hypothests, Re learney oearches are pace to Eaentiay the best hypotherts | fhe Earsget Ainchoo a Semple eementng cnctance: aa sino Hovag “roel “raske Paws Furr _eoleur Mowe: Sse. elghant No Shorts Yer isNlorn No Black No Be, Yer Yes Shet No Na Ne Brown Yor Medium Kio el Novetherh Yori yh ue Blickil Ne Medltum Yea al No__tong ib Yes_-Yex Wht no) fnedtum No 7 No Shorr Yes Ye Yer Black No Bes Yes >The tonset concept * +o fdentth the antmal tp 1an__ elephants _Explafn Concepe eee stasles St an elephant [fom the dataset GND taste. _atven data Pnpute & Pnstances each with & ath bu le : 1 Tas gee concept /Ainehon c+ Elephant > fiver No3 iHypoKerts Hz cet of hypothec% each %e congédchdn of teberale ac proposéhon Like, each tetera l- Ps _2epsecented ac” Alyebute > vatve ): The hy potherte for “the Concept leasntog of an element * geen oa pies reps h=< No short yer 9 9 Black 95 Thte hy pother® bh fs s f oN je loge ie ae below sia ie ae i (Hoe No) AlTaPl=shove) A (Tuske= Yeon (Por = 1) I Alfir =e) A (color = slack) A Comer = nloyA (stzes9)5 ERI) bolailadel IJ heoan hypothe?s h 40 predict beeen Such that dor a_agtven tert Tartanco » hi) =eCx) -Thés hypothect | Ge produced alo called as concept _decevépton which ff a Medel that can re used td ‘claesthy subsequent Pastances. Genevaligaton and spectltzatvon dew stai bout uw We _Conshiy ct beevare let us a wale, ci Kos s1elatfon. j uihy_genevaltgation of the most Spectife_hypothests frolws peehca bon of the mort general hypothert hipothert cpace Con be search dor ary appe™ ete that noatehe: all sve: Partances thats does hot natch, eink adve Fastance. nies ead =oean beng the faypotiaraa Spacer gh There ore tw9* Waye eh E (2a Specinlegakinn = geneza sto Specthics leavntng Genevalezalin = speche +0 sonal deasnine orede. : 9 ‘abie | Co 3 the taafning ie oo sn 0 and flluctyate spectise 12 ge neval learning dorm alle ov the mest © e most sieshetclfuer ; +ve fn pothesis: Dano Aig ThE Plluchxat x Bp G Put Raaeps (2 the mort specie hypoth oe noe ClacePed an fostanceTi: No shore Yer No No Black No Bis Yet l4ve Foctanes hir 9 fhe vs after observing alt the_tve fo stances a Appr? _hy parherte fs _ generated whtch can Now _clocetty £ any subsequent _+tve Fostanee to true. ~Classtcat Ond pdapive_ mu systems Classical! ine _syrtem has. co poner s | Popute bx OCe SS CURDUE Adoptive Leasnpng: suitem ealeiae | fusp.0! etna, “tabu gtu Gib Ae lee paral poe avaelabl __learn?n 4.DATE; PAGE: Leoaning Types UW Learn by memorization 2y Leon by Example Bhearq boy being +tBou } Leann by esttottoa Teas > Sef leaning - wie iaforceme: | et teasing te solve cee a PLeaaning by genercales 23209 explana Hon Desisn of a learning Taaintng Hlgortthm _% called leerentng -elgorttim nd | Syste that fe butt oeround a learntng_ Nefnvtons Focus of 4 steps: 1 Choosing O_t5.aFo% Teo thee 2» Detewmene a target funcHon togget Arnotfon Sy Repsesenta Hing tance) ; Spectltzalfon Genenaltgakon 40 spective ow Tht learning technsloqy well search: hypotheres sade fhe eee ome general _hypothedte =pace; a = No Short Yer No No Black No Medium Ver (vy ho = No ° ye 2 9 Black No~ Bi9> a Ty =< Nb [ong No Yes Yes whfte No medium Ale Go , Remove any _hypothects fncone sfebent wofth Desai) hy = 2681 Mer nh nt Birt Ps 9 Pe Ps eae oe SR Oo NON Gaara Ts eno Short Yer Yer Yer Black ..No Bf9. Yes hs fs __stmelag +o hy oT usihs. 3% the h ypothec& Space Senerated wWhtch wil classthy the posthve + : instance to tre ana rnesakve nstance :40 falee. Fila EBS N0i 2 Pin Bad? g2acdue? > nO) ADM Vets ic tay aDS Ge ae? Beale Ono Os Poise Qe ani Sut Nol, Op i Saveats 12? nak bass ae — | Efnd_¢ Blgoctthm: 1D __ Dp: postive Pnetancer fo fhe fratning data set Step 4: PneWaleze bh to the mest specie thes : Tea 2 General?=e the Freka! = ae Bve__fnotance. 4 £ach subs Son. 3 is a 2 nee % Tt 7 check fp: ° Je ebtonce aae ck for each alyibute va : thee ts . Bf the CthPbute value e -valye then do nothing, Hy potherts for_the ane ov om esi% Sift RoisDATE: Value SS PAGES £ "7 Negatve ange fe SoA — Aalecor efue rach Cal lenovo led. Oh (Zeca ae aa pic good iter | Seo cod fore] Sood Gdoa 5 ad g. Consides the trafntng datacet of a enstancer fn_table @ ft contatne the detatls Of pets of Stdent and thet _tekelyheod on gelff 4 en fenal semertes. Poly dnd S Be she cs 4: PniKaltze h +o the rnost spe =< 9 $410 pee Tie a: Genel the Sntal bape! ve Onstance +ve_Pnetance, $0 otherts h to tnolude___ PAGE: ae ch a: Simélagtky based learning Ot between fn) “An fnstance an enl tSafning data set. Instance based methods teas tobe] of a tect fs aT temper tnstance _ fs Mode! based _| data tv a 2Nearest Nete Ahour er = mesnahital-approachito similarly ent clacsteaky ee hiae ek) | srlocally” woatted egrescton sy learntng Vector quanksak | Selt _oargants, g-maps & Radfant bas fc __K_ meqrest inetshhoun whch Pe abniera eaten method used oy both classtieaton and neg guess psoblerns, 2 | © 2 f | (ORO) @& z Ne The visual Stepreseatatvon . of thes leanntng shovsn These ‘ ose two elacs oF obpects Ct andor Ghen Rey ven» test ?nstance T tRe chiemony the test Pnotance «2s idletosmined by looletng al the clas of k=3 Nearest neighbour. Thus the alact this tect Pnstance + fs treated as Tiatntng dabset ¥ GPA Assessment Paofeet_subroitted Reswtl 0 N+ [potas gs 3B Bs pv 80 il Pass | 3\fieg i e p 2h 6) 15 3 Shes 50 me & 8.2 Tas a 5 & ore 38 ‘ Spd tale 8S ait 3CONS AOR The SMACHE peatoe mance , a. Jet of 2 bata — Enstances Shou Phila | wohech _dlescother the peeformane Ley. : ot Fedsvey WDrident Pm a cCountve 4 @5pq Obtarney. py |_sernesteo, . Preys, The _ ?adepend j= the faxget vaotable fe s1ecuit cone he | dl@ecgete valid variable at taker tog a Pre * Boreas) he performance of “Sd clascPty «ohefhes the chident Pov © han Pa crcceg a Lahie, Given _a tect fnctame Tl61,40,5) and Set of Categories Tpacc, fat/3 alco called ae clacces ue need +40 use fhe tgatnitng Set to clareg test Pnstanae fnq_ _Euoltdran_ dectance . he tack OF clase¥fcaron” go to asso aca —t9.0 _ovbftacy gnetaneo keB Step 4; Calculate rm tance _hefwe test fnstance and each of Re ain? L@s shown Sey table tie. 5 re SL.NO |@ em Asseement cise -Rerult . Euoltdean Ab J esa. a5 8 Pass I Can-e. a = 80 Bo lpia | eh us so 38‘Iisrep 2+ S09 tithes clfetanes tri asd Greet aaelee ee __ the fast 3 nearest trafning ‘data fnstancs i the _hest thstances -the _sele hewn Fo table se __Bnstance gucltdtan Aéttance clas % 5-001 Fae) 5 to 05783 fat) a &:.022315, . Fael Hese 7 takes the 3 Nearest ests igs cae ac¢_ 4,5, 7 are cad Aectances: ‘Step 3: Predtetr the clase of The test Poskonse by mafor Voting + The ‘elas for the fest Prstances fs predéeted ac fat) Wetted Ffarplerny tuleeghted KANN ~ eigen Constder the same tyatnting data set § Use _wefghted KNN and detes mene the class _ Test Pystane t1-¢, 60,8) Find tdtan_ditgtance 3 GPA pissecsment _—ogubenined eiResutt EuateaPon aaeg fins gsSort the: dteronce: Pa accend?ng oxdes ond thace ya Sine Se select Avast _ Wengect Nefsh bor Toeranee ED class 5 10-92636 atl jes. 6 19-05653 asp part — 4 _ 18-3808) (Past) fotl cole 2 ‘step 9 Pra sock the” sof te _wetsnted voting _technigue — _ ___ compute fnverse of enc Is __Invevte dist in 0.06502 0.084 0-06 602 4 410. 4%05] fail ‘ Find cum of fnvegses sum = 0:06502" +0:08294 +0-06502 = 0124093 Computé the wetsht by’ aéveding each ?nvesse __ distance & : : ‘ Bnctonee ED Class qrvere ditt wetsht= TOs PE (o.8uss fot 0-06F02- 09 -3843u4- 2 P ni _paedick the: clase he: class a maxim — “Test fnstanee fe p09)EP oT” ~~ Neanest Centrotd class x [LY lelaseDATE PAGE: | _l Locally wleighted Reqresfon (LUI [Sino Salany —__Expendthve(?n howe al t 5 ea ee on i eb). See fl 4 t 8 iConstdew a stmple example with # Tnstancer™ sho. ‘ Hn table 4-10 and app! loca " wef hted weqrecto | . sof Using lean 1 on model acumen have omputed e i Py = 4-42 “B, = O62 Test _frstance_% =2 yl= hot Bix \ 724 OG2K2 3! = 5-90 ppl Neasert Ne&hbour mede! choose _k =3 closest fnstances L__ Shino |) X= Salary Expendehies oa EDs 3 sinlsiod | = as = V2 i Hetil 15 ers a 1 Instances 2,3 gy _d8stanoes. _The mean 9ATE: PAGE Hence the uetghts ef the clerose Fasea ces computed sas follous: ae bdefght of instance 23 Wa =e 22 Daumne oCesae SEI hoes aye =< © 2hOup : ~8.1aF oh ; =e. = 0-043 a fs called bandwidth pasameley =(4a-x)? ‘a | Weisht of fnetance 3 wa = € sho : = (2-2) 5 Rhy = @ 24ene =Gu= fi (ouaa)™ h = | taletght_of Tnstane 4 wy. e > = (roy Ao= 252 —0:6613 = O54 Fitted lfne dave sy feaoileuies bet _us mode) fe ee pa Y=Qot ae =the fitted l2ne for the Subs =I = 0:54 + 0-C6X x! — The Petod afta =O-54a eran fq Maleix fosen Xilweek) Jitsaler 2m thoura el 4 Efnd the (Pneay ego of the data product sales _& [email protected] [Snean maty?x_fosmSoj?- || Hese _clependent varfable’ x fs given’ as w= ft 2 Sp TAdependent vagfable yt =f1 3 4 #) Soe rey m =/ eee ye [YS eres / 4 A 4 vy 8 } 4 The Regressfon gtven AS Qe Oey Computabn_osdev th& _eqd 2 shown by clep by ce @ computation ot (x™x) tH/ ( fo 30) LComputarieon of (Tx)? Touiet LS SONG y } 7 (oS aoexe [ is = aed antervcept ae ~— SlopeXt Ye errs ie Mie eo eet) Te. 5 Bee To 90 aS 100 [e “110 nn 1 20 Constdes +00 Koch ftems Te, @ Ta whore aohal valoes 80, 15 “The Reg medel ps dict: the valve of feme Te,T7 as 715,85 -Find MAE , MSE RMSE l Relakve @msmE, CV.he well pass/fatlfe ee a” Le Entyopy Info(P) = Entropy InfolP).... . Py) =r(P,109,0P.) + Pa log, (Po)+----+ Prlogathy) DATE 3 _| Entropy: Enbropy fs the amount of uncestatnt| 23 standomness _ ____ Fandom _varstable event bro by Lower Entropy - lower _uncertatn? hegher cuncentarnety te v TH there are jo data Fartancer out Of 6 belongs to ive clage £ 4 eve claw fe entopy Ts _caletlated a8 Entro =[ 6 log, 6 4 (og 10 TOs To (0 Decfsfon Tree Introd Algorithme: ID3, CH-S, CART, CHATD , QUEST, GUEDE, CRUE, |_ cTREe ty Dp Btree conskuchon Real (Corn TOF €x6.3:| sl-No | cqpp | anter actvences rea ‘see II 29 | Ye vety g00d| good 1 | ze Alo good | moderale [va it = 24 No! pvs eo a 1 - ! [4 ; ae s 6 a | good SePA Bee Ci piastaloutht the entropy of the target clase offer ie 15 we a " a oe bed Bs istesatson da ote 9; Caleulate eS atletbute ta the. trofatog date set ag Entropy intoomation i Copp | Job offer = Yes | Job offer = Noj Total® a zy 3 t He ee ze 4 o a a Gui as ze 0 oz [}DATE for practtca| knowledge ‘i dob offer=Yes_| Job oHer =No_| Total Fic OF 636) Gao = o- 8%o7-0-63¢) =O:2uy6 _fntrepy Pofosmatten for _communtaakon skille Ror Corre urese 50 |e raeeetes Job offer =No | ro1ah [eotrep Seed ee i 5 _toderale = ze z Peso posi 2 He | r ; Entropy Sof ("T, communteakon leis) 7 | || | Tl || I TiIpnterachveness | ctiveness | OO eee Prackeal know | O€@h6 | = | 2 entropy Ps __rntofenum And__.'. Gata the best _splet_altvtbule- >The best Split abkibute 0 : >So we choose coph ac 4h not values 4Enkopy Dnfotr) = Entropy anfol 3.) — Yeres Noe} oe 3 ! 1 {Sto9_3 +tsg. tb) = = (-O3nIt -0-499y = 0-108 Enkropy tnfo(T_,Pntevactiveness) Thlerackveners | Joboffer=ver |Joboffer=NO Lroral [én Hopy Ye ol | e20 4 [3g No BR 7 FE: ez [re oR ig, Soles 2) 4 Sf tog fly! “i le 0.4997 (Gain = 09108 - 0.4997 = 0-3iI) [Procigtl_knowledae = 6.8108 Communttakton slefllr= 0.9108 “Hencecu.5 _constructon Deane ee cus 2s an mprovement ove D3. cys Works with confoucut and dAtecres olfpi and: Mersin valves and tt aleo suppor | pyuntng. i _Step 4 :Caleutate clase @nfopy fox target el Make use of — nfowmaho APN _ OF whech cge__caloulated rps _algoréthm @n e to construct a Aec?rfon tree using CH Enbro| ri Tasgor -pttbute = 30b offen) = Eokopy _ nfo, 3) legit ce Gtk oe Je iS] = 2019013 sagen apis ain = 08807-03243 £0. 2 ak Splet Info (1, CG PA) = = an Rake_ Reqrecifon Taees. Assecement _~Ase&gnment aaig oe Seed oe. Ste 38 Pv3 Yes sewer TS Gaiters 01 agit nae aes iin Good Yay Mey a9¢ Dug No Ya go 7 Good a No Le | g Poor Yel Yes fr 4 ps no No 5g to Gooat ves ye a9 Si eS IE+NS FIP FHOt I FOS tS tES 10 Sai anda Oe fakon = fo IPH (10 = 1S on LE TE co | = 16.55 _Step ai Dsseement = good | SLNO_ Assessment Ass’gnment Profect Recut i leat good Yes ve idie | ms 900d No Yes qc [3 good Yes SC ae, ha good No No 1S | 5 good, Yee Yeo 84 | heh al. fvg = FST PEt tICHE cg 86.4 s