Ukuqala iphrojekthi entsha yokufunda ngomshini kuletha intshiseko esheshayo, futhi kungase kulingeke ngempela ukugxumela ekugcineni ekujuleni. Kunamamodeli akamuva aphambili noma ama-algorithms ayinkimbinkimbi okungenzeka ukuthi ufunde ngawo. Bathembisa imiphumela ephusile, futhi ukugwema isilingo sokuzama ngabo ngokushesha nje ebhethini kuwumsebenzi onzima.  Noma yimuphi usomabhizinisi wesimanje umagange ukuhlola amasu asezingeni eliphezulu futhi abonise amaphrojekthi ayinkimbinkimbi (nayimpumelelo) emphakathini. Nokho, lo mdlandla, nakuba umuhle, ngezinye izikhathi ungathatha isikhathi esibalulekile njengoba ulungisa kahle ama-hyperparameter futhi uhlangabezane nobunzima bokusebenzisa amamodeli ayinkimbinkimbi.  Kule nqubo, kunombuzo owodwa obalulekile okufanele ubuzwe:  Sikukala kanjani ngempela ukusebenza kahle kwemodeli yethu?  Ukuthola ukuthi inkimbinkimbi yemodeli yethu ifanelekile yini noma uma ukusebenza kuphakeme ngokweqiniso kungaba yinselele. Lokhu kwenzeka uma lingekho   Lapha, ukuba nemodeli yesisekelo kubaluleke kakhulu. Isisekelo sinikeza lelo phuzu lereferensi elibalulekile - liqondile, liyashesha ukwakha, futhi liyachazeka ngokwemvelo. Ngokumangalisayo, ngokuvamile imodeli eyisisekelo, engathatha kuphela u-10% wesamba somzamo wokuthuthukisa, ingakwazi ukufeza kufikela ku-90% womsebenzi oyifunayo, ikhiqize indlela esebenza kahle kakhulu yemiphumela ephusile. iphuzu elilula lokubhekisela kulo.  Umqondo wokuqala okulula akuyona nje indlela elula yabaqalayo - kuwumkhuba obalulekile ohlala ubalulekile kuzo zonke izigaba zomsebenzi wesayensi yedatha. Kuwuhlelo oluyisisekelo kanye nesikhumbuzo esihle sokulinganisela isifiso sethu sokuba yinkimbinkimbi nezinto ezisebenzisekayo zezixazululo ezicacile, eziqondakala kalula, nezilawulekayo.  Ukuqonda Amamodeli Ayisisekelo  Imodeli yesisekelo   esetshenziselwa ukubhekana nenkinga. Ngokuvamile, lawa mamodeli afaka ukuhlehla komugqa emiphumeleni eqhubekayo noma ukuhlehla kokuhleleka kwemiphumela yezigaba. Isibonelo, ukuhlehla komugqa kungabikezela ukubuyiselwa kwesitoko ngokusekelwe kudatha yentengo yomlando, kuyilapho ukwehla kwezinto kungase kuhlukanise abafake izicelo zekhredithi njengengcuphe ephezulu noma ephansi. iyinguqulo eyisisekelo  Le ndlela ihlukile kumamodeli ayinkimbinkimbi kakhulu njengamanethiwekhi e-neural noma izindlela ezihlanganisayo, okuthi, nakuba zinamandla, zingenza ukubamba inkinga kube nzima kakhulu futhi kwandise isikhathi esidingekayo sokuthuthukiswa ngenxa yobunkimbinkimbi bazo nezinsiza ezibalulekile zokubala.   Izinzuzo Zokuqala Ngemodeli Eyisisekelo   Ukulinganisa izilinganiso  Ukulinganisa izilinganiso kuyisinyathelo sokuqala esibaluleke kakhulu ekuthuthukisweni kwanoma iyiphi imodeli ye-ML. Uma usetha imodeli yesisekelo, usungula imethrikhi yokusebenza eyisisekelo lapho wonke amamodeli alandelayo (okuvame ukuba yinkimbinkimbi kakhulu) kufanele adlule ukuze aqinisekise ubunkimbinkimbi bawo kanye nokusetshenziswa kwensiza. Le nqubo ayikona nje ukuhlola okuhle kwengqondo kodwa futhi isekela lokho okulindele futhi ikunikeza isilinganiso esicacile senqubekelaphambili.  Isibonelo, cabanga uthuthukisa imodeli yokubikezela izitayela zemakethe yezezimali usebenzisa i-avareji ehambayo elula (SMA) njengesisekelo. Le SMA ingase isebenzise idatha yomlando yesikhathi esifushane ukuze ibikezele izintengo zesitoko zesikhathi esizayo, izuze ukunemba kokuqala okungama-60% ekubikezeleni ukunyakaza kwezimakethe ngendlela efanele. Le modeli ibe isisetha ibhentshimakhi yanoma imaphi amamodeli athuthukile alandelayo. Uma imodeli eyinkimbinkimbi, njengenethiwekhi Yenkumbulo Yesikhathi Esifushane (LSM), ithuthukiswa kamuva futhi ifinyelela ukunemba okungu-65%, ukukhuphuka kokusebenza kungalinganiswa ngokunembile ngokuqhathaniswa nesisekelo sokuqala esingu-60%.  Lokhu kuqhathanisa kubalulekile ekunqumeni ukuthi ukuthuthukiswa kokunemba okungu-5% kuyabuthethelela yini ubunkimbinkimbi obengeziwe kanye nezidingo zokubala ze-LSTM. Ngaphandle kwesisekelo esinjengalesi, ukwenza izinqumo ezinolwazi mayelana nokukaleka nokusebenzisa okungokoqobo kwamamodeli ayinkimbinkimbi kuba yinselele.  Le ndlela yokulinganisa yenza isiqiniseko sokuthi ukuthuthukiswa kobunzima bemodeli kuyafaneleka futhi kuzoholela ekuthuthukisweni kwangempela, konke lokhu kwenza inqubo yokuthuthukisa ihambisane nemiphumela ephumelelayo.   Izindleko-Ngempumelelo  Ukulandela indlela eyongayo ku-ML kubalulekile. Ikakhulukazi uma uzibekela umgomo wokuvumelanisa izinqubo zakho nezimiso ezibeka   Uma uqala ngemodeli eyisisekelo, wehlisa izinsiza nesikhathi esidingekayo ekuthuthukisweni nasekuhloleni imodeli yokuqala. Lokhu kusho ukwenza i-prototyping esheshayo - futhi lokho kubalulekile ukuze uthole impendulo esheshayo nentuthuko ephindaphindwayo. inani elikhulu ngenkathi unciphisa ukumosha njengento ebaluleke kakhulu.  Ngalesi sisekelo, noma iyiphi inkimbinkimbi oyengezayo manje ingahlolwa ngokucophelela.  Isibonelo, uma ufuna ukwenza ukushintshela ku-algorithm eyinkimbinkimbi efana ne-vector autoregression (VAR) futhi uthole ukuthi ikhuphula kancane ukunemba kokubikezela, udinga ukucabanga kabusha ukuthi ingabe lokhu kuthuthukiswa okuncane kuyazithethelela yini izimfuno zekhompyutha ezengeziwe nobunkimbinkimbi. Impendulo ingase ibe cha. Khona-ke imodeli elula ihlala inketho engabizi kakhulu.  Ngokugxila ekusebenziseni izindleko, uqinisekisa ukuthi izinsiza zisetshenziswa kahle futhi uzuza okungaphezu nje kwezithuthukisi zobuchwepheshe. Futhi, iletha izixazululo ezingokoqobo, ezenezelelwe inani elifanele mayelana nokuthuthukiswa kokusebenza nokwabiwa kwezinsiza. Ngale ndlela, ukutshalwa kwezimali ngakunye kobunzima bemodeli kuqinisekisiwe, okufaka isandla emigomeni yonkana yephrojekthi ngaphandle kwezindleko ezingaphandle kwesilinganiso.   Ukungafihli kanye nokutolika  Emikhakheni efana nezezimali lapho izinqumo kufanele zihambisane nezindinganiso eziqinile zokulawula, ukucaca kwamamodeli akuyona nje inzuzo yebhizinisi. Kuyindlela yesu esiza kakhulu ohlelweni lokuhlangana nemithethonqubo futhi kusiza ukuxhumana kalula nababambiqhaza okungenzeka ukuthi abanaso isizinda (esijulile) sobuchwepheshe.  Ake sithathe imodeli yethu ye-SMA. Ihunyushwa kalula ngoba imiphumela yayo ihlobene ngokuqondile nedatha yokufaka. Lokhu kwenza kube lula ukuchaza ukuthi okokufaka ngakunye kuwuthonya kanjani umphumela obikezelwe. Uma izinqumo ezisuselwe kuzibikezelo zemodeli zidinga ukuthethelelwa kubalawuli bangaphandle noma ngaphakathi kumalungu eqembu angewona ochwepheshe, lobu lula buwukhiye ezinqubweni zakho.  Uma isinqumo esisuselwe kuzibikezelo zemodeli ye-SMA singatshazwa, ukucaca kwemodeli kuvumela incazelo esheshayo nelula yomqondo ongemuva komsebenzi wayo. Lokhu kungasiza ngezibuyekezo zokulawula nokuhlola futhi kuthuthukise ukwethembana nokutholwa phakathi kwabasebenzisi nabathatha izinqumo. Ngaphezu kwalokho, njengoba ubunkimbinkimbi bemodeli bukhula, ngokwesibonelo ukuthuthela kuma-algorithms ayinkimbinkimbi afana namamodeli e-ARIMA noma e-VAR ukuze uthole izibikezelo ezinokwethenjelwa, ukutolika kwesisekelo sokuqala se-SMA kuba uphawu lokubonisa ukuthi yiliphi izinga lencazelo okudingeka ulethule.  Ngokusebenzisa ama-regressor afana nesici sokubaluleka amaphuzu noma amanani e-SHAP ahlanganiswe namamodeli ayinkimbinkimbi, ukuqhubeka kwanoma iyiphi enye indlela yokusebenza kwemodeli kuhlala kusobala. Lokhu kusiza inhloso yenqubo yokuphepha ukuthi ingalahlwa kumamodeli athuthuke kakhulu. Iphuzu lemodeli yesisekelo esilula ukusebenzisa njalo umbandela wokuthi ukwakheka okuphelele nokubaluleka kuzogcinwa ngisho noma izinga lobunzima likhula. Lokhu kuqinisekisa izinhlinzeko zokuthobela kanye nokuxhumana okuzosebenza ngempumelelo.   Ukulawulwa Kwengozi  Ukulawulwa kobungozi kungenye into ebalulekile yokuthuthukisa amamodeli okufunda emishini, ikakhulukazi emikhakheni efana nezezimali lapho izibikezelo ezinembile nezithembekile zinomthelela ekuthathweni kwezinqumo. Ukuba nemodeli yesisekelo esilula kuyisu elihle lokulawula lezi zingozi.  Isisekelo esiqondile sinikeza isiqalo esiqondakalayo, esikuvumela ukuthi kancane kancane (futhi ngokuphephile) wengeze izithuthukisi kubunkimbinkimbi bemodeli.  Isibonelo, imodeli ye-SMA (yize iyisisekelo) yenza isisekelo esiqinile sokuthola amaphethini ayisisekelo kanye nokungahambi kahle okungenzeka ekunyakazeni kwentengo yesitoko. Ukuyisebenzisa kusiza ukukhomba izimpawu zokuqala zokuntengantenga noma ukuziphatha okungavamile kwemakethe. Ukwenza lokho kubalulekile, ukugwema ubungozi bezezimali ngaphambi kokusebenzisa ama-algorithms wokuqagela ayinkimbinkimbi.  Ngaphezu kwalokho, ukusebenzisa imodeli eyisisekelo kunciphisa ubungozi bokugcwala ngokweqile. Kuwumgodi ovamile ekumodeleni kwezezimali. Ukufakwa ngokweqile kwenzeka uma imodeli ishunwe kahle kakhulu kudatha yomlando futhi ithwebula umsindo kunephethini eyisisekelo. Ngenxa yalokhu, ungathola izibikezelo ezidukisayo futhi uthole amasu okuhweba angathembekile ngenxa yalokho. Imodeli elula enamapharamitha ambalwa ayithambekele kangako kule nkinga, iqinisekisa ukuthi ukuqagela ekunikezayo ngokuvamile kusebenza kudatha engabonakali.  Ukwandisa ubunkimbinkimbi njengoba i-SMA ithuthukela kumodeli encane ehambayo efana ne-ARIMA ne-VAR iba inkimbinkimbi, isakhiwo esilula se-SMA singasisiza sicabangele ngokuhlelekile ukusebenza kwenkimbinkimbi ngayinye eyengeziwe. Lokhu kuthuthukiswa okuhamba kancane kokuyinkimbinkimbi kusiza ukugcina ukulawula ukusebenza kwemodeli, iqinisekisa ukuthi isendlalelo ngasinye esiyinkimbinkimbi esingeziwe sinikeza inzuzo ecacile futhi asilethi ingozi engadingekile.  Le ndlela ehlelekile yokukhula kwemodeli eyinkimbinkimbi isiza ekuqondeni ukuthi izinguquko kumodeli zithinta kanjani ukuziphatha nokuthembeka kwayo. Iphinde iqinisekise ukuthi ubungozi buhlala bulawulwa kahle. Uma uqala ngesisekelo esilula futhi ulawula ngokucophelela isigaba ngasinye sokuthuthuka, uyaqinisekisa ukuthi amamodeli okubikezela ahlala enamandla futhi ephephile, esekela izinqumo zezezimali.   Ukucatshangelwa Okubalulekile Lapho Usebenzisa Amamodeli Ayisisekelo  Ukuze ukhethe imodeli eyisisekelo efaneleke kakhulu, udinga ukuqonda inkinga yebhizinisi nezici zedatha. Isibonelo, izibikezelo zochungechunge lwesikhathi zezimakethe zezimali zingase ziqale ngemodeli ye-ARIMA njengesisekelo sokuthwebula amandla esikhashana ngendlela elula. Ikhwalithi yedatha nokucubungula kusengaphambili nakho kudlala indima ebalulekile; ngisho nemodeli elula ingasebenza kabi uma inikezwa idatha enganele noma engacutshungulwanga kahle ngaphambili.  Futhi okokugcina, ukwazi ukuthi kufanele ushintshe nini ukusuka kwesisekelo ukuya kumodeli eyinkimbinkimbi kubalulekile. Lesi sinqumo kufanele siqondiswe ukuhlolwa okukhulayo nokuqinisekiswa, ngokuhambisana nendlela ye-Agile yokuphindaphinda.   Ngamafuphi  Ukuqala amaphrojekthi akho okufunda ngomshini ngokwethula imodeli yesisekelo esilula akusona nje isinyathelo sokuqala. Kuyisu. Isu elihambisana nezindlela ze-Agile ezithuthukisa ukusebenza kahle, ukusebenza kahle, nokuvumelana nezimo. Ukusondela kuphrojekthi yakho ngale ndlela kungathuthukisa kakhulu imiphumela yephrojekthi ngokuqinisekisa ukuthi konke ukwanda kobunzima kuyafaneleka futhi kwengeza inani elibonakalayo. Ukwamukela ubulula kuyinto enamandla. Kuyisu elihle kakhulu emikhakheni efana nezezimali lapho izinqumo kufanele zisheshe.

Arweave

Read My Stories

Lo msindo ukhiqizwa ngolimi lokuqala lwendaba!

Kude kakhulu; Uzofunda

Make resilience your competitive advantage

Ukuqala Okulula: Inzuzo Yamasu Yamamodeli Asesisekelo Ekufundeni Ngomshini

About Author

IMIBONO

HANG TAGS

LESI SIHLOKO SETHULWE NGAPHAKATHI

Related Stories

The Billionaires F*cked Up

Zulu Network: Moving The Bitcoin Economy Forward With a Two-Tiered Bitcoin Layer 2 Architecture

The Billionaires F*cked Up

Zulu Network: Moving The Bitcoin Economy Forward With a Two-Tiered Bitcoin Layer 2 Architecture

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps