Isiqulatho[Fihla][Bonisa]
Ihlabathi litshintsha ngokukhawuleza ngenxa yobukrelekrele bokwenziwa, kunye nokufunda koomatshini, okunempembelelo kuzo zonke iinkalo zobomi bethu bemihla ngemihla.
Ukusuka kubancedisi belizwi abasebenzisa i-NLP kunye nokufunda koomatshini ukubhukisha ukuqeshwa, ukujonga iziganeko kwikhalenda yethu, kwaye udlale umculo ukuya kwizixhobo ezichanekileyo kangangokuba banokulindela iimfuno zethu ngaphambi kokuba siziqwalasele.
Iikhompyuter zinokudlala ichess, zenze utyando, kwaye ziphuhliseke zibe ngoomatshini abakrelekrele, abafana nabantu ngoncedo lweealgorithms zokufunda koomatshini.
Sikwixesha lenkqubela phambili yetekhnoloji, kwaye ngokubona indlela iikhompyuter eziphuhliswe ngayo ngokuhamba kwexesha, sinokwenza uqikelelo malunga nokuya kwenzeka kwixesha elizayo.
Idemokhrasi yezixhobo zekhompyutha kunye neendlela ngomnye wemiba ephambili yolu tshintsho olubalaseleyo. Iinkcukacha zesayensi baye benza iikhomputha ezinamandla zokucola idatha kule minyaka mihlanu idlulileyo ngokuphumeza ngaphandle kokuzama iindlela zokusika. Iziphumo ziyamangalisa.
Kule post, siza kujonga ngokusondeleyo yokufunda umatshini I-algorithms kunye nazo zonke iintlobo zazo.
Ke, zithini iialgorithms zokuFunda ngoMatshini?
Indlela esetyenziswa yinkqubo ye-AI ukuphumeza umsebenzi wayo-ngokubanzi, ukubikezela amaxabiso okuphuma kwidatha enikiweyo-yaziwa ngokuba yi-algorithm yokufunda ngomatshini.
I-algorithm yokufunda ngomatshini yinkqubo esebenzisa idatha kwaye isetyenziselwa ukwenza iimodeli zokufunda ngomatshini ezilungele ukuveliswa. Ukuba ukufunda ngoomatshini nguloliwe owenza umsebenzi, ngoko ke iialgorithms zokufunda ngoomatshini ziziporo ezihambisa umsebenzi kunye.
Eyona ndlela ibalaseleyo yokufunda koomatshini oza kuyisebenzisa iya kumiselwa yingxaki yeshishini ozama ukuyilungisa, uhlobo lwedatha oyisebenzisayo, kunye nezixhobo onazo.
Ii-algorithms zokufunda ngoomatshini zezo zijika iseti yedatha ibe yimodeli. Kuxhomekeka kuhlobo lwengxaki ozama ukuyiphendula, amandla okusebenza akhoyo, kunye nohlobo lwedatha onayo, egadiweyo, engajongwanga, okanye i-algorithms yokufunda yokuqinisa inokusebenza kakuhle.
Ke, sithethe malunga nokufunda okugadwayo, okungagadwayo, kunye nokuqiniswa, kodwa zithini? Makhe sizihlolisise.
UkuFundisa okuBekelwe iliso, okungagadwanga kunye nokomelezwa
UkuFunda ngokuBekwa
Kwimfundo ephantsi kolawulo, imodeli ye-AI iphuhliswa ngokusekelwe kwigalelo elinikiweyo kunye neleyibhile emele isiphumo esiqikelelweyo. Ngokusekwe kumagalelo kunye neziphumo, imodeli iphuhlisa i-equation yemephu, kwaye isebenzisa loo equation yemephu, iqikelela ileyibhile yamagalelo kwixesha elizayo.
Masithi kufuneka senze imodeli ekwazi ukwahlula phakathi kwenja nekati. Iifoto ezininzi zeekati kunye nezinja zondliwa kwimodeli eneeleyibhile ezibonisa ukuba ziikati okanye izinja ukuze kuqeqeshwe imodeli.
Imodeli izama ukuseka i-equation enxulumene neelebhile kwiifoto zegalelo kuloo mifanekiso. Nangona imodeli ayizange ibone umfanekiso ngaphambili, emva koqeqesho, inokuchonga ukuba ingaba yikati okanye inja.
UkuFunda okungalawulwa
Ukufunda okungajongwanga kubandakanya ukuqeqesha imodeli ye-AI kuphela kumagalelo ngaphandle kokuleyibhile. Imodeli yahlula idatha yegalelo ibe ngamaqela aneempawu ezinxulumeneyo.
Ileyibhile yexesha elizayo yegalelo ke ngoko iqikelelo ngokuxhomekeka kwindlela iimpawu zayo ezisondelelene ngayo nolunye ulwahlulo. Cinga ngemeko apho kufuneka sahlule iqela leebhola ezibomvu neziluhlaza zibe ngamacandelo amabini.
Makhe sicinge ukuba ezinye iimpawu zebhola ziyafana, ngaphandle kombala. Ngesiseko sendlela enokwahlula ngayo iibhola kwiiklasi ezimbini, imodeli ibheka iimpawu ezahlukileyo phakathi kweebhola.
Amaqela amabini eebhola-enye eluhlaza okwesibhakabhaka kunye nebomvu-iveliswa xa iibhola zihlulwe zibe ngamaqela amabini ngokusekelwe kwi-hue yazo.
Ukomeleza ukuFunda
Ekufundeni okomeleza, imodeli ye-AI ifuna ukwandisa inzuzo iyonke ngokusebenza njengoko inakho kwimeko ethile. Ingxelo ngeziphumo zayo zangaphambili inceda imodeli ifunde.
Cinga ngemeko xa irobhothi iyalelwe ukuba ikhethe indlela phakathi kwamanqaku A no-B. Irobhothi ikhetha kuqala nokuba yeyiphi na yezifundo kuba ayinamava angaphambili.
Irobhothi ifumana igalelo kwindlela eyithathayo kwaye ifumana ulwazi kuyo. Irobhothi inokusebenzisa igalelo ukulungisa umba kwixesha elizayo xa idibana nemeko efanayo.
Umzekelo, ukuba irobhothi ikhetha ukhetho B kwaye ifumana umvuzo, njengengxelo entle, iyaqonda ngeli xesha ukuba kufuneka ikhethe indlela B ukunyusa umvuzo wayo.
Ngoku ekugqibeleni into eniyilindeleyo nonke, zii-algorithms.
Ii-algorithms zokuFunda koomatshini abakhulu
1. Ukuhlehla ngomgca
Eyona ndlela ilula yokufunda ngoomatshini etenxayo kwimfundo ebekwe iliso kukuhlehla ngomgca. Ngolwazi oluvela kwiinguqu ezizimeleyo, lusetyenziswa kakhulu ukusombulula imiba yokuhlehla kunye nokudala uqikelelo kwizinto ezixhomekeke rhoqo.
Ukufumana umgca weyona nto ifanelekileyo, enokunceda ekuqikeleleni isiphumo sezinto eziqhubekayo ezixhomekeke kuzo, yinjongo yokuguqulwa komgca. Amaxabiso ezindlu, ubudala, kunye nemivuzo yeminye yemizekelo yamaxabiso aqhubekayo.
Imodeli eyaziwa ngokuba yi-linear regression elula isebenzisa umgca othe ngqo ukubala unxulumano phakathi kolunye uguquko oluzimeleyo kunye nolunye oluxhomekeke. Kukho ngaphezu kwezibini ezizimeleyo eziguquguqukayo kuluhlu oluphindaphindiweyo lwelayini.
Imodeli yokubuyisela umgca ineengcinga ezine ezisisiseko:
- Umgca: Kukho umdibaniso womgca phakathi kuka-X kunye nentsingiselo ka-Y.
- I-Homoscedasticity: Ngexabiso ngalinye lika-X, umahluko oshiyekileyo uyafana.
- Ukuzimela: Uqwalaselo luzimele omnye komnye ngokwemiqathango yokuzimela.
- Ukuqheleka: Xa u-X elungisiwe, uY uqhele ukusasazwa.
Ubuyiselo lomgca lusebenza ngokuncomekayo kwidatha enokwahlulwa ngokwemigca. Inokulawula ukugqwesa ngokugqithisileyo ngokusebenzisa uhlengahlengiso, ukuqinisekiswa okunqamlezayo, kunye neendlela zokunciphisa ubukhulu. Nangona kunjalo, kukho iimeko apho ubunjineli obubanzi bufuneka, obunokuthi ngamanye amaxesha bubangele ukugqithisa kunye nengxolo.
2. ULungiselelo loLungiselelo
Ukuhlehla ngomatshini yenye indlela yokufunda koomatshini eshiya ukufunda okugadwayo. Olona setyenziso lwayo luphambili luhlelo, ngelixa inokusetyenziselwa iingxaki zokubuyela umva.
Uhlengahlengiso lolungiselelo lusetyenziselwa ukuqikelela ukuguquguquka okuxhomekeke ngokwecandelo kusetyenziswa ulwazi olusuka kwizinto ezizimeleyo. Injongo kukuhlela iziphumo, ezinokuwela kuphela phakathi kwe-0 kunye ne-1.
Itotali elinganisiweyo yamagalelo iqhutywe ngumsebenzi we-sigmoid, umsebenzi wokuvuselela oguqula ixabiso phakathi kwe-0 kunye ne-1.
Isiseko sokuhlehliswa kolungiselelo luqikelelo lokunokwenzeka okukhulu, indlela yokubala iiparamitha zonikezelo olucingelwayo olunikiweyo olujongwe idatha ethile.
3. Umthi wesigqibo
Enye indlela yokufunda ngomatshini eqhekekayo kwimfundo egadiweyo ngumthi wesigqibo. Kuyo yomibini imiba yokuhlelwa kunye nokuhlehla, indlela yomthi wesigqibo ingasetyenziswa.
Esi sixhobo sokwenza izigqibo, esifana nomthi, sisebenzisa imiboniso ebonakalayo ukubonisa iziphumo ezilindelekileyo zezenzo, iindleko, kunye neziphumo. Ngokwahlula idatha ibe ngamacandelo ahlukeneyo, ingcamango ifana nengqondo yomntu.
Idatha yahlulwe yangamacandelo ahlukeneyo kangangoko sinokuyigranula. Eyona njongo yoMthi weSigqibo kukwakha imodeli yoqeqesho enokuthi isetyenziswe ukuqikelela udidi loguquko ekujoliswe kulo. Amaxabiso alahlekileyo anokuphathwa ngokuzenzekelayo usebenzisa uMthi weSigqibo.
Akukho mfuneko ye-encoding ye-shot-encoding, i-dummy variables, okanye amanye amanyathelo onyango lwangaphambili lwedatha. Iqinile ngengqiqo yokuba kunzima ukongeza idatha entsha kuyo. Ukuba unedatha eyongezelelweyo enombhalo, kufuneka uphinde uqeqeshe umthi kuyo yonke idataset.
Ngenxa yoko, imithi yesigqibo lukhetho olubi kuyo nayiphi na isicelo esifuna utshintsho oluguquguqukayo lwemodeli.
Ngokusekelwe kuhlobo loguquko ekujoliswe kulo, imithi yesigqibo ihlelwa ngokweendidi ezimbini:
- I-Categorical Variable: Umthi weSigqibo apho injongo iguquguquka ngokweCategorical.
- Uguquguquko oluqhubekayo: Umthi weSigqibo apho ukuguquguquka kwenjongo kuQhubeka.
4. Random Forest
I-Random Forest Method yindlela elandelayo yokufunda umatshini kwaye i-algorithm yokufunda yomatshini egadiweyo esetyenziswa ngokubanzi kwimiba yokuhlelwa kunye nokubuyisela umva. Ikwayindlela esekwe emthini, efana nomthi wesigqibo.
Ihlathi lemithi, okanye imithi emininzi yesigqibo, isetyenziswa ngendlela yehlathi engacwangciswanga ukwenza izigwebo. Xa kusingathwa imisebenzi yokuhlelwa, indlela yehlathi engacwangciswanga isebenzise iiguquguquki zecategorical ngelixa kusingathwa imisebenzi yohlengahlengiso kunye neeseti zedatha eziqulethe izinto eziguquguqukayo eziqhubekayo.
Ukudityaniswa, okanye ukuxutywa kweemodeli ezininzi, yinto eyenziwa yindlela yehlathi engacwangciswanga, nto leyo ethetha ukuba uqikelelo lwenziwa kusetyenziswa iqela leemodeli kunokuba libe linye.
Ikhono lokusetyenziselwa zombini ukuhlelwa kunye neengxaki zokuhlehla, ezenza uninzi lweenkqubo zokufunda zoomatshini zanamhlanje, luyinzuzo ephambili yehlathi elingaqhelekanga.
Iindlela ezimbini ezahlukeneyo zisetyenziswa yi-Ensemble:
- I-Bagging: Ngokwenza oku, iinkcukacha ezininzi ziveliswa kwidathasethi yoqeqesho. Ukunciphisa ukuhluka koqikelelo, oku kuyenziwa.
- I-Boosting yinkqubo yokudibanisa abafundi ababuthathaka kunye nabafundi abomeleleyo ngokwakha iimodeli ezilandelelanayo, okukhokelela kwimodeli yokugqibela echanekileyo kakhulu.
5. Naive Bayes
I-binary (iiklasi ezimbini) kunye nomcimbi wokuhlelwa kweeklasi ezininzi unokusombululwa ngokusebenzisa ubuchule beNaive Bayes. Xa indlela ichazwa kusetyenziswa amaxabiso okubini okanye odidi lwegalelo, kulula ukuyiqonda. Ingqikelelo eyenziwe ngumdidiyeli weNaive Bayes kukuba ubukho belinye inqaku eklasini abunanxaxheba kubukho bazo naziphi na ezinye iimpawu.
Le fomula ingentla ibonisa:
- P (H): Amathuba okuba i-hypothesis H ichanekile. Okunokwenzeka kwangaphambili kubhekiselwa kuko ngolu hlobo.
- P (E): Ukubakho kobungqina
- P (E | H): Ubunokwenzeka bokuba i-hypothesis ixhaswa bubungqina.
- P (H | E): Amathuba okuba i-hypothesis iyinyani, inikwe ubungqina.
Umdidiyeli weNaive Bayes unokuthathela ingqalelo nganye yezi mpawu xa emisela ukubakho kwesiphumo esithile, nokuba ezi mpawu ziqhagamshelwe kwenye. Imodeli yeNaive Bayesian ilula ukuyakha kwaye iyasebenza kwiiseti zedatha ezinkulu.
Kwaziwa ngokusebenza bhetele nangaphezu kwezona ndlela zintsonkothileyo zokuhlela ngelixa isisiseko. Yingqokelela ye-algorithms zonke zisekwe kwiTheorem yeBayes, kunendlela enye.
6. K-Abamelwane abakufutshane
Ubuchule babamelwane be-K (i-kNN) liqela elisezantsi eliphantsi kweliso lomatshini lokufunda elinokusetyenziselwa ukujongana nemiba yokuhlelwa kunye nohlengahlengiso. I-algorithm ye-KNN iqikelela ukuba izinto ezinokuthelekiswa zinokufunyanwa kufutshane.
Ndiyikhumbula njengendibano yabantu abanengqondo efanayo. I-kNN yenza inzuzo yombono wokufana phakathi kwezinye iindawo zedatha usebenzisa ukusondela, ukusondela, okanye umgama. Ukuze uleyibhelishe idatha engabonakaliyo ngokusekwe kweyona ndawo ikufutshane ephawulwe ngeenkcukacha zedatha enokubonwa, kusetyenziswa indlela yemathematika ukumisela ukwahlulwa phakathi kwamanqaku kwigrafu.
Kufuneka umisele umgama phakathi kwamanqaku edatha ukuze uchonge ezona ndawo zikufutshane ezinokuthelekiswa. Imilinganiselo yomgama enjengomgama we-Euclidean, umgama weHaming, Manhattan umgama, kunye nomgama weMinkowski ungasetyenziselwa oku. U-K waziwa njengelona nani lommelwane likufutshane, kwaye lisoloko lilinani elingumnqakathi.
I-KNN ingasetyenziswa kwiingxaki zokuhlelwa kunye nokubuyisela umva. Uqikelelo olwenziwe xa i-KNN isetyenziselwa ukubuyisela imiba isekelwe kwintsingiselo okanye i-median ye-K-iziganeko ezifanayo.
Isiphumo se-algorithm yohlelo olusekwe kwi-KNN inokumiselwa njengeklasi ene-frequency ephezulu phakathi kwe-K yezehlo ezifanayo. Imeko nganye ivota kudidi lwabo, kwaye uqikelelo lolodidi olufumana ezona voti zininzi.
7. K-indlela
Bubuchule bokufunda obungajongwanga obujongana nemiba yokuhlanganisana. Iiseti zedatha zahlulahlulwe ngokwenani elithile lamaqela-fowuna masiyenze i-K-ngendlela yokuba amanqaku edatha yeqela ngalinye ahluke kwaye ahluke kwezo zikwamanye amaqela.
K-ithetha indlela yokudibanisa:
- Kwiqela ngalinye, i-algorithm ye-K ikhetha i-k centroids, okanye amanqaku.
- Ngezona centroids ezikufutshane okanye amaqela e-K, indawo nganye yedatha yenza iqoqo.
- Ngoku, iicentroids ezintsha ziveliswa ngokuxhomekeke kumalungu eqela asele ekhona.
- Owona mgama ukufutshane kwindawo nganye yedatha ubalwa kusetyenziswa ezi centroids zihlaziyiweyo. Ukude i-centroids ingatshintshi, le nkqubo iphinda iphindwe.
Ikhawuleza, ithembekile, kwaye kulula ukuyiqonda. Ukuba kukho imiba, i-k-means' adaptability yenza uhlengahlengiso lube lula. Xa iiseti zedatha zahlukile okanye zahlulwe kakuhle enye kwenye, iziphumo zingcono. Ayinakulawula idatha engalunganga okanye ngaphandle.
8. Ukuxhasa oomatshini beVector
Xa usebenzisa ubuchule be-SVM ukuhlela idatha, idatha ekrwada iboniswa njengamachaphaza kwindawo engu-n-dimensional (apho u-n linani leempawu onazo). Idatha emva koko inokuhlelwa ngokulula kuba ixabiso lefitsha ngalinye lidityaniswe kulungelelwaniso oluthile.
Ukwahlula idatha kwaye uyibeke kwigrafu, sebenzisa imigca eyaziwa ngokuba ngabahluli. Le ndlela icwangcisa inqaku ledatha nganye njengenqaku kwindawo engu-n-dimensional, apho u-n linani leempawu onazo kunye nexabiso lenqaku ngalinye lixabiso elithile lokulungelelanisa.
Ngoku siza kufumana umgca owahlula-hlula idatha kwiiseti ezimbini zedatha ezibekwe ngokwahlukileyo. Imigama ukusuka kwiindawo ezikufutshane kwiqela ngalinye kula mabini iya kuba yeyona iqeleleneyo kulo mgca.
Ekubeni amanqaku amabini asondeleyo awona akude kakhulu kumgca kumzekelo ongentla, umgca owahlula idatha kumaqela amabini ahlelwe ngokwahlukileyo ngumgca ophakathi. Umdidi wethu ngulo mgca.
9. UkuNcitshiswa koMda
Ukusebenzisa indlela yokunciphisa i-dimensionality, idatha yoqeqesho ingaba neenguqu ezimbalwa zegalelo. Ngamagama alula, ibhekisa kwinkqubo yokucutha ubungakanani beseti yenqaku lakho. Masicinge ukuba iseti yedatha yakho ineekholamu ezili-100; ukuncitshiswa kobukhulu kuya kuncipha loo mali ukuya kwikholamu ezingama-20.
Imodeli ikhula ngokuzenzekelayo ngakumbi kwaye inomngcipheko omkhulu wokugqithisa njengoko inani leempawu likhuphuka. Owona mcimbi mkhulu ngokusebenza ngedatha kwimilinganiselo emikhulu yinto eyaziwa ngokuba "sisiqalekiso sobukhulu," eyenzeka xa idatha yakho iqulethe inani eligqithisileyo leempawu.
Ezi zinto zilandelayo zingasetyenziselwa ukufezekisa ukunciphisa ubukhulu:
- Ukufumana kunye nokukhetha iimpawu ezifanelekileyo, ukhetho lweempawu luyasetyenziswa.
- Ukusebenzisa iimpawu esele zikhona, inqaku lobunjineli lidala izinto ezintsha.
isiphelo
Ukufunda ngoomatshini okungajongwanga okanye okugadiweyo kuyenzeka zombini. Khetha ukufunda okugadwayo ukuba idata yakho ininzi kakhulu kwaye ithegiwe kakuhle kuqeqesho.
Iiseti ezinkulu zedatha zihlala zisebenza kwaye zivelise iziphumo ezingcono zisebenzisa ukufunda okungajongwanga. U kufunda o lukhulu iindlela zingcono ukuba unengqokelela yedatha enkulu efumaneka lula.
Ukuqinisa ukufunda kunye nokufunda ukomeleza okunzulu zezinye zezihloko ozifundileyo. Iimpawu zothungelwano lweNeural, ukusetyenziswa, kunye nemithintelo ngoku icacile kuwe. Okokugqibela kodwa okungakuncinananga, uqwalasele iinketho kwiilwimi ezahlukeneyo zokucwangcisa, ii-IDE, kunye namaqonga xa kuziwa ekudaleni eyakho. iimodeli zokufunda ngomatshini.
Into elandelayo ekufuneka uyenzile kukuqalisa ukufunda nokusebenzisa nganye yokufunda umatshini ukusondela. Nokuba isifundo sibanzi, nasiphi na isihloko sinokuqondwa kwiiyure ezimbalwa ukuba ugxininise kubunzulu baso. Isifundo ngasinye sizimele sodwa kwezinye.
Kufuneka ucinge ngomba ube mnye ngexesha, uwufunde, uwusebenzise, kwaye usebenzise ulwimi olukhethileyo ukuphumeza i(ii)algorithm kuyo.
Shiya iMpendulo