Elikufutshane indlela ummelwane: umzekelo umsebenzi

indlela ummelwane sisondeleyo somhleli elula metric osekelwe phezu kuvavanyo yokufana bezinto ezahlukeneyo.

into ngokuhlalutywa bobabo iklasi ukuba bengabakho ngabalawulwa sample yoqeqesho. Makhe sibone ukuba yeyiphi na ummelwane ikufuphi. Zama ukuqonda lo mbandela inzima, imizekelo iindlela ezahlukeneyo.

indlela hypothesis

elikufutshane indlela ummelwane nga ozakuthathwa algorithm ixhaphakileyo esetyenziselwa yokuhlela. Into usisisulu yokuhlela omelwe ukwenziwa y_i eklasini, apho into ekufutshane ukufunda isampula x_i.

Ingakumbi iindlela elikufutshane abamelwane

k elikufutshane ummelwane Lendlela ukuphucula ngokuchaneka yokuhlela. into ngokuhlalutywa bobabo kwiklasi enye njengokuba isambuku kubamelwane ayo, oko kukuthi, k kusondela kuyo izinto le x_i isampuli zihlalutywe. Ekucombululeni iingxaki kunye neeklasi ezimbini inani kubamelwane kuya kuba odd ukuphepha imeko ambiguity, ukuba inani elifanayo kubamelwane ziya kunikwa kwiiklasi ezahlukeneyo.

Ubuchule abamelwane emsebenzini

indlela PostgreSQL-zihlalutywe tsvector abamelwane elikufutshane kusetyenziswa xa inani leeklasi ubuncinane ezithathu, kwaye awukwazi ukusebenzisa inani limnqakathi. Kodwa kuvela ukungacaci nokuba kwezi meko. Emva koko, lo ummelwane i-th ufumana ubunzima w_i, leyo kunciphisa akudidi umakhelwane i. Libhekisela ukuba iklasi le into, nto leyo eya kuba ubunzima obuyi phakathi kubamelwane abasenyongweni.

Le hypothesis of compactness

Xa intliziyo zonke iindlela ngentla hypothesis of compactness. Icebisa ukuba unxulumano phakathi umlinganiselo yokufana lezinto obuphezu kwabo eklasini enye. Kulo meko, umda phakathi kweentlobo ezahlukeneyo luhlobo elula, kwaye ukudala kwiiklasi yezinto isithuba kwindawo compact mobile. Phantsi iindawo ezinjalo kuhlalutyo zemathematika ithathwa ukuba ibhekisele iseti ivaliwe iphahlwe. Le hypothesis akazalani imbono yemihla ngemihla zelizwi.

Le ndlela yokubala basic

Makhe sihlolisise ezinye ummelwane elikufutshane. Ukuba uqeqesho ecetywayo uhlobo isampuli 'into-impendulo »X ^ m = \ {(x_1, y_1), \ amachaphaza, (x_m, y_m) \}; ukuba ezininzi izinto ukuchaza umgama umsebenzi \ rho (x, x '), nto leyo limelwe ngokohlobo imodeli ukufana olwaneleyo bezinto ngokwandisa ixabiso umsebenzi liyancipha ukufana phakathi kwezinto x, x'.

Malunga nayo nayiphi na into, u uya kwakha isampuli yoqeqesho izinto x_i kunye ukwandisa imigama ukuba u:

\ Rho (u, x_ {1; u}) \ leq \ rho (u, x_ {2; u}) \ leq \ cdots \ leq \ rho (u, x_ {m; u}),

apho x_ {i; u} ephawula isampulu yokufunda into, i-th ummelwane source into u. notation nokusetyenziswa olunjalo ukuphendula ummelwane i-th: y_ {i; u}. Ngenxa yoko, sifumanisa ukuba nasiphi na isixhobo u Uxhokonxwa nokunikezwa isampula yabo.

Ukubekwa inombolo k abamelwane

elikufutshane indlela ummelwane xa k = 1 iyakwazi enika yokuhlela ngempazamo, hayi kuphela izinto-ukukhutshwa, kodwa ezinye iiklasi ekufuphi.

Xa sithatha k = m, i algorithm iya kuba kuzinza kwaye uya idlulele ekubeni ixabiso rhoqo. Kungenxa yoko le nto ukuthembeka kubalulekile ukuphepha indices kakhulu k.

Ukuziqhelisa, kwingongoma nganye njengoko isalathiso k eliphezulu kusetyenziswa sliding ulawulo.

ezikhutshwayo luhlola

Iinjongo zokufunda ubukhulu becala engalinganiyo, kodwa phakathi kwazo kukho abo iimpawu kwiklasi yaye kuthiwa nemigangatho. Xa kufutshane kuxhomekeke imodeli ifanelekileyo lilonke yayo ephezulu yelo eklasini.

Indlela rezultativen indlela abamelwane ekufutshane? Umzekelo kungabonwa ngokusekelwe iindidi peripheral and non-ulwazi yezinto. Kucingelwa bume oxineneyo into abanye abameli kule klasi. Xa uyisuse lucalulo kweesampula umgangatho ayisivumeli.

Fumana kwinani elithile leesampuli bangase Ungqabhuko ingxolo ukuba 'emhlabeni' ukuba iklasi. Ukususa negalelo kakhulu elungileyo umgangatho yokuhlela.

Ukuba isampuli ethatyathiweyo kulo izinto ingxolo uninformative nokuphelisa, unako ukubala phezu iziphumo embalwa entle ngexesha elifanayo.

Lokuqala indlela zelwe ye ulwahlulohlulo ummelwane elikufutshane ivumela ukuphucula umgangatho, ukunciphisa ubungakanani data egcinwe, ukunciphisa ixesha yokuhlelwa, nto leyo ichithwa ezikhethiweyo imigangatho elandelayo.

Ukusetyenziswa iisampuli ultra-enkulu

elikufutshane indlela ummelwane isekelwe phezu yokugcina yokwenene izinto zokufunda. Ukudala kakhulu iisampula ezinkulu-isikali usebenzisa ingxaki yezobuchwepheshe. Injongo nje ukuze ugcine isixa esikhulu solwazi, kodwa umlinganiselo ubuncinane ixesha nexesha ukufumana nayiphi na into u k phakathi abamelwane basondeleyo.

Ukumelana nalo msebenzi, iindlela ezimbini ezisetyenziswayo:

• isampuli thinned nge izinto obhobhozayo non-data;
• ukusetyenziswa nesakhiwo sedata olusebenzayo ezizodwa kunye neekhowudi sophendlo ngephanyazo abamelwane ekufutshane.

Imithetho le ndlela yokukhetha

Ulwahlulo zingentla ingqalelo. Elikufutshane indlela ummelwane isetyenziswa ekucombululeni iingxaki ezibonakalayo, nto leyo eyaziwa kwangaphambili umgama umsebenzi \ rho (x, x '). Xa izinto echaza zithwala yamanani basebenzisa metric Euclidean. Le choice akukho isizathu ezikhethekileyo, kodwa kubandakanya umlinganiselo zonke iimpawu "kumanqanaba efanayo." Ukuba le nto ithe kuthathelwa ingqalelo, ngoko ke engenayo iya ngabona phawu ukuba aphezulu amaxabiso yamanani.

Ukuba kukho isixa esikhulu iimpawu, ukubala ibanga njengoko sum of zokwahluka kwi iimpawu ezithile zibonakala kakhulu ubukhulu ingxaki.

Xa indawo eliphezulu mbini kude ukusuka omnye kwesinye zonke izinto. Ekugqibeleni, nayiphi na isampuli iya kuba ecaleni into efundwayo abamelwane k. ukhethe inani elincinane iimpawu efundisayo ukuphelisa le ngxaki. Ubuchule zokubala iingqikelelo ekwakheni ngokusekelwe leeseti ezahlukeneyo imiqondiso, kwaye umntu ngamnye ukwakha umsebenzi yabo isondele.

isiphelo

izibalo kuvamile ukusetyenziswa iintlobo ubuchule ukuba babe zabo neempawu ezikhethekileyo, ezilungileyo nezingalunganga. Kujongwa indlela ummelwane elikufutshane bebenoyilungisa kakhulu yingxaki enkulu, ngenxa iimpawu zezinto zezibalo. Ingcamango yokulinga, ngokusekelwe indlela yacazululwa eyenziwayo ngenkuthalo ukusetyenziswa basemoyeni.

Xa iinkqubo isazi i kuyimfuneko nje ukuba lokuhlela izinto, kodwa bonisa umsebenzisi ingcaciso ulwahlulohlulo kuthethwa. Kule ndlela, ingcaciso le nto ichazwa ngokunxulumene into kwiklasi ethile ngokunjalo indawo kuthelekiswa yayo lesampula kusetyenziswa. iingcali kushishino Legal, zokuma, amagqirha, thatha le "ophambili" logic basebenzise ngenkuthalo oko kuphando lwabo.

Ukuze zihlalutywe ndlela yaba inokuthenjwa, eyiyo, enika iziphumo ezinqwenelekayo, kufuneka uthathe umzobo k ubuncinane, lo gama kwakhona ukuphepha ezikhutshwayo phakathi izinto zihlalutywe. Kungenxa yoko le nto ukusetyenziswa kwemigangatho kunye indlela yokhetho, kwakunye metric kahle.