Okuqukethwe[Fihla][Bonisa]
Umhlaba ushintsha ngokushesha ngenxa yobuhlakani bokwenziwa, nokufunda komshini, okunomthelela kuzo zonke izici zokuphila kwethu kwansuku zonke.
Kusukela kubasizi bezwi abasebenzisa i-NLP nokufunda komshini ukuze babhukhe ama-aphoyintimenti, babheke imicimbi ekhalendeni lethu, futhi badlale umculo kuye kumadivayisi anembe kangangokuthi angalindela izidingo zethu ngaphambi kokuthi sizicabangele.
Amakhompyutha angadlala i-chess, enze ukuhlinzwa, futhi athuthuke abe imishini ehlakaniphe kakhulu, efana nomuntu ngosizo lwama-algorithms okufunda komshini.
Sisesikhathini sokuthuthuka okuqhubekayo kwezobuchwepheshe, futhi ngokubona indlela amakhompyutha athuthuke ngayo ngokuhamba kwesikhathi, singenza izibikezelo mayelana nokuzokwenzeka esikhathini esizayo.
Ukwenziwa ngentando yeningi kwamathuluzi nezindlela zekhompiyutha kungenye yezinto ezibalulekile zalolu guquko olugqamayo. Ososayensi bedatha badale amakhompuyutha anamandla okuhlanganisa idatha phakathi neminyaka emihlanu edlule ngokusebenzisa kalula izindlela ezisezingeni eliphezulu. Imiphumela iyamangalisa.
Kulokhu okuthunyelwe, sizobheka eduze ukufunda imishini ama-algorithms nakho konke ukuhlukahluka kwawo.
Ngakho-ke, ayini ama-algorithms wokufunda ngomshini?
Indlela esetshenziswa uhlelo lwe-AI ukwenza umsebenzi wayo—ngokuvamile, ukubikezela amanani okukhiphayo kusuka kudatha yokufaka enikeziwe—yaziwa ngokuthi i-algorithm yokufunda komshini.
I-algorithm yokufunda komshini iyinqubo esebenzisa idatha futhi isetshenziselwa ukudala amamodeli okufunda omshini alungele ukukhiqizwa. Uma ukufunda komshini kuyisitimela esenza umsebenzi, khona-ke ama-algorithms wokufunda komshini yizitimela ezihambisa umsebenzi.
Indlela engcono kakhulu yokufunda umshini ongayisebenzisa izonqunywa inkinga yebhizinisi ozama ukuyilungisa, uhlobo lwedathasethi oyisebenzisayo, nezinsiza onazo.
Ama-algorithms wokufunda komshini yilawo aguqula isethi yedatha ibe yimodeli. Kuye ngohlobo lwenkinga ozama ukuyiphendula, amandla okucubungula atholakalayo, kanye nohlobo lwedatha onalo, ama-algorithms wokufunda agadiwe, angagadiwe, noma okuqinisa angenza kahle.
Ngakho-ke, sikhulume ngokufunda okugadiwe, okungagadiwe, nokuqinisa, kodwa kuyini lokho? Ake sizihlole.
Ukufunda Okugadiwe, Okungagadiwe Nokuqinisa
Ukufunda Okugadiwe
Ekufundeni okugadiwe, imodeli ye-AI ithuthukiswa ngokusekelwe kokokufaka okunikeziwe kanye nelebula elimele umphumela obikezelwe. Ngokusekelwe kokufakwayo nokuphumayo, imodeli ithuthukisa isibalo semephu, futhi isebenzisa leyo sibalo semephu, ibikezela ilebula lokokufaka esikhathini esizayo.
Ake sithi sidinga ukwakha imodeli ekwazi ukuhlukanisa phakathi kwenja nekati. Izithombe eziningi zamakati nezinja zifakwa kumodeli enamalebula abonisa ukuthi amakati noma izinja ukuze kuqeqeshwe imodeli.
Imodeli ifuna ukusungula isibalo esihlobene namalebula ezithombeni ezifakiwe kulezo zithombe. Ngisho noma imodeli ingakaze ibone isithombe ngaphambili, ngemva kokuqeqeshwa, ingakwazi ukubona ukuthi ikati noma inja.
Ukufunda okungagadiwe
Ukufunda okungagadiwe kubandakanya ukuqeqesha imodeli ye-AI kuphela kokufakwayo ngaphandle kokuyilebula. Imodeli ihlukanisa idatha yokokufaka ngamaqembu anezici ezihlobene.
Ilebula yesikhathi esizayo yokokufaka ibe isibikezelo kuye ngokuthi izibaluli zayo zifana ngokuseduze kangakanani nesigaba esisodwa. Cabanga ngesimo lapho kufanele sihlukanise khona iqembu lamabhola abomvu naluhlaza abe yizigaba ezimbili.
Ake sicabange ukuthi ezinye izici zamabhola ziyefana, ngaphandle kombala. Ngesisekelo sokuthi ingahlukanisa kanjani amabhola abe amakilasi amabili, imodeli ibheka izici ezihlukile phakathi kwamabhola.
Amaqoqo amabili amabhola-elilodwa eluhlaza okwesibhakabhaka nelilodwa elibomvu-akhiqizwa lapho amabhola ehlukaniswa ngamaqembu amabili ngokusekelwe kumbala wawo.
Ukuqinisa Ukufunda
Emfundweni yokuqinisa, imodeli ye-AI ifuna ukukhulisa inzuzo iyonke ngokwenza ngendlela engenza ngayo esimeni esithile. Impendulo ngemiphumela yayo yangaphambili isiza imodeli ukuthi ifunde.
Cabanga ngesimo lapho irobhothi liyalwa ukuba likhethe umzila phakathi kwamaphoyinti A no-B. Irobhothi liqale likhethe noma yisiphi isifundo ngoba alinaso isipiliyoni sangaphambilini.
Irobhothi lithola okokufaka emzileni eliwuthathayo futhi lizuza ulwazi kulo. Irobhothi lingasebenzisa okokufaka ukuze lilungise inkinga ngokuzayo lapho lihlangabezana nesimo esifanayo.
Isibonelo, uma irobhothi likhetha inketho B futhi lithola umklomelo, njengempendulo eyakhayo, manje liyaqonda ukuthi kufanele likhethe indlela B ukuze lenyuse umvuzo walo.
Manje ekugcineni enikulindile nonke, ama-algorithms.
Ama-algorithms Wokufunda Womshini Omkhulu
1. Ukwehla Komugqa
Indlela elula yokufunda yomshini ephambuka ekufundeni okugadiwe ukuhlehla komugqa. Ngolwazi oluvela kokuguquguqukayo okuzimele, lusetshenziswa kakhulu ukuxazulula izinkinga zokuhlehla nokudala ukuqagela kokuhluka okuncike okuqhubekayo.
Ukuthola umugqa wokulingana okungcono kakhulu, okungasiza ekubikezeleni umphumela wokuhlukahluka okuncikile okuqhubekayo, inhloso yokuhlehla komugqa. Izintengo zezindlu, ubudala, namaholo ezinye zezibonelo zamanani aqhubekayo.
Imodeli eyaziwa ngokuthi ukuhlehla komugqa okulula isebenzisa umugqa oqondile ukubala ukuhlotshaniswa phakathi kokuhluka okukodwa okuhlukile nokukodwa okuncikile okukodwa. Kukhona okuhlukile okuzimele okungaphezu kokubili ekuhlehleni komugqa okuningi.
Imodeli yomugqa wokuhlehla inemibono emine eyisisekelo:
- I-Linearity: Kunokuxhumana komugqa phakathi kuka-X kanye nencazelo ka-Y.
- I-Homoscedasticity: Kuwo wonke amanani ka-X, ukuhluka okuyinsalela kuyefana.
- Ukuzimela: Okubonwayo kuzimele komunye nomunye mayelana nokuzimela.
- Okujwayelekile: Uma u-X elungisiwe, u-Y uvame ukusatshalaliswa.
Ukwehla komugqa kusebenza kahle kudatha engahlukaniswa ngemigqa. Ingakwazi ukulawula ukufaka ngokweqile ngokusebenzisa ukujwayela, ukuqinisekiswa okuphambene, kanye namasu okunciphisa ubukhulu. Kodwa-ke, kunezimo lapho ubunjiniyela besici esibanzi budingeka, okungase ngezinye izikhathi kubangele ukugcwala ngokweqile nomsindo.
2. Logistic Regression
Ukuhlehla kokuhamba kungenye indlela yokufunda yomshini ephuma ekufundeni okugadiwe. Ukusetshenziswa kwayo okukhulu ukuhlukanisa, kuyilapho ingasetshenziswa futhi ezinkingeni zokuhlehla.
Ukuhlehla kwezinto kusetshenziswa ukubikezela ukuhluka okuncikile kwesigaba kusetshenziswa ulwazi oluvela kuzinto ezizimele. Umgomo uwukuhlukanisa okuphumayo, okungaba kuphela phakathi kuka-0 no-1.
Isamba esinesisindo sokufakwayo sicutshungulwa umsebenzi we-sigmoid, umsebenzi wokwenza kusebenze oguqula amanani phakathi kuka-0 no-1.
Isisekelo sokuhlehla kwezinto siwukulinganisa okukhulu kwamathuba, indlela yokubala amapharamitha wokusatshalaliswa kwamathuba okucatshangwayo kunikezwe idatha ethile ephawuliwe.
3. Isihlahla Sesinqumo
Enye indlela yokufunda yomshini ehlukana phakathi kokufunda okugadiwe isihlahla sesinqumo. Kuzo zombili izindaba zokuhlukaniswa nokwehla, indlela yesihlahla sesinqumo ingasetshenziswa.
Leli thuluzi lokuthatha izinqumo, elifana nesihlahla, lisebenzisa izethulo ezibonakalayo ukuze libonise imiphumela elindelekile yezenzo, izindleko, kanye nemiphumela. Ngokuhlukanisa idatha ibe izingxenye ezihlukene, umqondo ufana nomqondo womuntu.
Idatha ihlukaniswe yaba izingxenye ezihlukene kangangoba besingakwazi ukuyihlanganisa. Inhloso enkulu yeSihlahla Sesinqumo ukwakha imodeli yokuqeqesha engasetshenziswa ukubikezela ikilasi lokuguquguquka okuhlosiwe. Amanani angekho angasingathwa ngokuzenzakalelayo kusetshenziswa Isihlahla Sesinqumo.
Asikho isidingo sombhalo wekhodi weshothi elilodwa, okuguquguqukayo okuyidumi, noma ezinye izinyathelo zokwelashwa kusengaphambili kwedatha. Iqinile ngomqondo wokuthi kunzima ukwengeza idatha entsha kuyo. Uma uthole idatha eyengeziwe enamalebula, kufanele uqeqeshe kabusha isihlahla kuyo yonke idathasethi.
Ngenxa yalokho, izihlahla zesinqumo ziyinketho engalungile yanoma yiluphi uhlelo lokusebenza oludinga ukuguqulwa kwemodeli eguquguqukayo.
Ngokusekelwe kuhlobo lokuguquguquka okuhlosiwe, izihlahla zokunquma zihlukaniswa zibe izinhlobo ezimbili:
- I-Categorical Variable: Isihlahla Sesinqumo lapho okuguquguqukayo kwegoli Kungokwesigaba.
- Okuguquguqukayo Okuguquguqukayo: Isihlahla Sesinqumo lapho okuguquguqukayo kwegoli Kuqhubekayo.
4. Ihlathi Elingahleliwe
I-Random Forest Method iyindlela yokufunda yomshini elandelayo futhi iyi-algorithm yokufunda yomshini egadiwe esetshenziswa kakhulu ezinkingeni zokuhlela nokuhlehla. Futhi kuyindlela esekelwe esihlahleni, efana nesihlahla sokunquma.
Ihlathi lezihlahla, noma izihlahla eziningi zokunquma, kusetshenziswa indlela yehlathi engahleliwe ukwenza izahlulelo. Lapho usingatha imisebenzi yokuhlukanisa, indlela yehlathi engahleliwe isebenzise okuguquguqukayo kwezigaba ngenkathi isingatha imisebenzi yokuhlehla ngamadathasethi aqukethe okuguquguqukayo okuqhubekayo.
Iqoqo, noma ukuhlanganiswa kwamamodeli amaningi, yilokho okwenziwa indlela yehlathi engahleliwe, okusho ukuthi izibikezelo zenziwa kusetshenziswa iqoqo lamamodeli kuneyodwa nje.
Ikhono lokusetshenziselwa kokubili izinkinga zokuhlukanisa nezokuhlehla, ezakha iningi lezinhlelo zokufunda zemishini yesimanje, kuyinzuzo eyinhloko yehlathi elingahleliwe.
Amasu amabili ahlukene asetshenziswa yi-Ensemble:
- Ukupakisha: Ngokwenza lokhu, idatha eyengeziwe ikhiqizwa kudathasethi yokuqeqeshwa. Ukuze kuncishiswe ukuhlukahluka kwezibikezelo, lokhu kuyenziwa.
- I-Boosting iyinqubo yokuhlanganisa abafundi ababuthakathaka nabafundi abaqinile ngokwakha amamodeli alandelanayo, okuholela kumodeli yokugcina enembayo enkulu.
5. Naive Bayes
Inkinga yokuhlukanisa kanambambili (yezigaba ezimbili) kanye nezigaba eziningi ingaxazululwa kusetshenziswa indlela ye-Naive Bayes. Uma indlela ichazwa kusetshenziswa amanani okufaka kanambambili noma esigaba, kulula ukuyiqonda. Umcabango owenziwe i-Naive Bayes classifier ukuthi ubukhona besici esisodwa ekilasini abuthinti ukuba khona kwanoma yiziphi ezinye izici.
Ifomula engenhla ikhombisa:
- P (H): Amathuba okuthi i-hypothesis H ilungile. Amathuba angaphambili abizwa ngokuthi yilokhu.
- P (E): Amathuba obufakazi
- P(E|H): Amathuba okuthi i-hypothesis isekelwa ubufakazi.
- P (H|E): Amathuba okuthi i-hypothesis iyiqiniso, uma kunikezwe ubufakazi.
Isigaba se-Naive Bayes sizocabangela isici ngasinye salezi zimpawu ngazinye lapho sinquma ukuba nokwenzeka komphumela othile, ngisho noma lezi zibaluli zixhumene kwesinye. Imodeli ye-Naive Bayesian ilula ukwakha futhi isebenza kahle kumadathasethi amakhulu.
Kwaziwa ngokwenza kangcono kunamasu okuhlukanisa ayinkimbinkimbi kuyilapho iyisisekelo. Kuyiqoqo lama-algorithms wonke asekelwe ku-Bayes' Theorem, esikhundleni sendlela eyodwa.
6. K-Omakhelwane Abaseduze
Indlela yomakhelwane be-K abaseduze (i-kNN) isethi engaphansi yokufunda komshini ogadiwe engasetshenziswa ukuze kubhekwane nezinkinga zokuhlela nokuhlehla. I-algorithm ye-KNN ithatha ngokuthi izinto ezifanayo zingatholwa eduze.
Ngikukhumbula njengomhlangano wabantu abanomqondo ofanayo. I-kNN yenza ithuba lomqondo wokufana phakathi kwamanye amaphoyinti edatha kusetshenziswa ukusondela, ukusondelana, noma ibanga. Ukuze kufakwe ilebula kudatha engabonakali ngokusekelwe kumaphuzu edatha aseduze anelebula abonakalayo, kusetshenziswa indlela yezibalo ukuze kunqunywe ukuhlukaniswa phakathi kwamaphoyinti kugrafu.
Kufanele unqume ibanga eliphakathi kwamaphoyinti edatha ukuze ukhombe izindawo ezifanayo eziseduze. Izilinganiso zebanga njengebanga le-Euclidean, ibanga le-Hamming, ibanga le-Manhattan, nebanga le-Minkowski lingasetshenziswa kulokhu. I-K yaziwa njengenombolo yomakhelwane eseduze, futhi ngokuvamile iyinombolo eyinqaba.
I-KNN ingasetshenziswa ezinkingeni zokuhlela nokuhlehla. Isibikezelo esenziwa lapho i-KNN isetshenziselwa izinkinga zokuhlehla sisekelwe kuncazelo noma i-median yezenzeko ezifanayo kakhulu zika-K.
Umphumela we-algorithm yokuhlukanisa ngokusekelwe ku-KNN unganqunywa njengekilasi elinemvamisa ephezulu kakhulu phakathi kwezenzeko zika-K ezifanayo kakhulu. Sonke isikhathi sivotela isigaba sabo, futhi ukubikezela kungokwesigaba esithola amavoti amaningi.
7. K-izindlela
Kuyisu lokufunda okungagadiwe elibhekana nezinkinga zokuhlanganisa. Amasethi edatha ahlukaniswe abe inombolo ethile yamaqoqo—sishayele sithi masiyenze K—ngendlela yokuthi amaphuzu edatha yeqoqo ngalinye ahluke futhi ahluke kulawo akwezinye iziqoqo.
K-isho indlela yokuhlanganisa:
- Kuqoqo ngalinye, i-algorithm ye-K-means ikhetha ama-k centroids, noma amaphoyinti.
- Ngama-centroid aseduze noma amaqoqo ka-K, iphoyinti ledatha ngalinye lakha iqoqo.
- Manje, ama-centroids amasha akhiqizwa ngokuya ngamalungu eqoqo asevele akhona.
- Ibanga eliseduze kakhulu lephoyinti ledatha ngalinye libalwa kusetshenziswa lawa ma-centroid abuyekeziwe. Kuze kube yilapho ama-centroids engashintshi, le nqubo iyaphindwa.
Iyashesha, ithembekile, futhi kulula ukuyiqonda. Uma kunezinkinga, ukuzivumelanisa nezimo kwe-k-means kwenza ukulungisa kube lula. Uma amasethi edatha ehlukile noma ehlukanisiwe kahle ukusuka kwelinye, imiphumela iba ngcono kakhulu. Ayikwazi ukuphatha idatha engalungile noma izinto eziphuma ngaphandle.
8. Imishini yokusekela iVector
Uma usebenzisa indlela ye-SVM ukuze uhlukanise idatha, idatha eluhlaza iboniswa njengamachashazi esikhaleni esingu-n-dimensional (lapho u-n kuyinombolo yezici onazo). Idatha bese ihlukaniswa kalula ngenxa yokuthi inani lesici ngasinye lisuke lixhunywe kusixhumanisi esithile.
Ukuze uhlukanise idatha futhi uyibeke kugrafu, sebenzisa imigqa eyaziwa ngokuthi abahlukanisi. Le ndlela ihlela iphoyinti ledatha ngalinye njengephoyinti endaweni engu-n-dimensional, lapho u-n eyinombolo yezici onazo futhi inani lesici ngasinye liyinani elithile lokuxhumanisa.
Manje sizothola umugqa ohlukanisa idatha kumasethi amabili edatha ahlelwe ngokwehlukana. Amabanga ukusuka emaphoyinti aseduze eqenjini ngalinye kulawa amabili azoqhelelana kakhulu kulo mugqa.
Njengoba amaphuzu amabili aseduze kakhulu yiwo aqhelelene kakhulu nomugqa osesibonelweni esingenhla, umugqa ohlukanisa idatha emaqenjini amabili ahlukaniswe ngokwehlukana umugqa ophakathi. Isihlukanisi sethu yilo mugqa.
9. Ukuncishiswa kobukhulu
Ngokusebenzisa indlela yokwehliswa kobukhulu, idatha yokuqeqeshwa ingase ibe nokuhlukahluka kokokufaka okumbalwa. Ngamagama alula, isho inqubo yokunciphisa usayizi wesethi yesici sakho. Ake sicabange ukuthi idathasethi yakho inamakholomu angu-100; ukuncishiswa kobukhulu kuzokwehlisa lelo nani libe amakholomu angu-20.
Imodeli ikhula ngokuzenzakalelayo ibe yinkimbinkimbi futhi inengozi enkulu yokugcwala ngokweqile njengoba inani lezici likhuphuka. Inkinga enkulu ngokusebenza nedatha ngobukhulu obukhulu yilokho okwaziwa ngokuthi "isiqalekiso sobukhulu," okwenzeka lapho idatha yakho iqukethe inani eliningi lezici.
Izinto ezilandelayo zingasetshenziswa ukufeza ukunciphisa ubukhulu:
- Ukuthola nokukhetha izici ezifanele, kusetshenziswa ukukhethwa kwezici.
- Ngokusebenzisa izici ezikhona kakade, isici sobunjiniyela sidala mathupha izici ezintsha.
Isiphetho
Ukufunda komshini okungagadiwe noma okugadiwe kuyenzeka kokubili. Khetha ukufunda okugadiwe uma idatha yakho incane futhi imakwe kahle ukuze uqeqeshwe.
Amasethi edatha amakhulu ngokuvamile angenza futhi akhiqize imiphumela engcono esebenzisa ukufunda okungagadiwe. Ukufunda okujulile izindlela zingcono kakhulu uma uneqoqo ledatha elikhulu elitholakala kalula.
Ukuqiniswa kokufunda nokufunda okujulile kokuqinisa ngezinye zezihloko ozifundile. Izici, ukusetshenziswa, kanye nemikhawulo yamanethiwekhi e-Neural manje sekucacile kuwe. Okokugcina, ucabangele izinketho zezilimi ezihlukene zokuhlela, ama-IDE, nezinkundla uma kuziwa ekudaleni owakho. amamodeli wokufunda wemishini.
Okulandelayo okudingeka ukwenze ukuqala ukufunda nokusebenzisa ngayinye ukufunda imishini sondela. Ngisho noma isihloko sibanzi, noma yisiphi isihloko singaqondwa emahoreni ambalwa uma ugxila ekujuleni kwaso. Isihloko ngasinye sizimele sodwa kwezinye.
Kufanele ucabange ngodaba olulodwa ngesikhathi, ulufunde, ulusebenzise, futhi usebenzise ulimi oluthandayo ukuze usebenzise i(ama)algorithm ekuyo.
shiya impendulo