Muna 2018, iyo pfungwa yeMLOps yakaonekwa mumakwikwi ehunyanzvi uye pamisangano yemusoro yakatsaurirwa kuAI, iyo yakakurumidza kubata muindasitiri uye yave kusimudzira segwara rakazvimirira. Mune ramangwana, MLOps inogona kuve imwe yenzvimbo dzakakurumbira muIT. Chii uye chinodyiwa nei tinzwe pazasi.
Chii chinonzi MLOps
MLOps (kusanganisa matekinoroji ekudzidza muchina uye maitiro uye nzira dzekushandisa dzakagadziridzwa mamodheru kuita bhizinesi maitiro) inzira nyowani yekubatana pakati pevamiriri vebhizinesi, masayendisiti, masvomhu, nyanzvi dzekudzidza muchina uye mainjiniya eIT pakugadzira masisitimu ehungwaru.
Mune mamwe mazwi, inzira yekushandura nzira dzekudzidza muchina uye matekinoroji kuita chishandiso chinobatsira chekugadzirisa matambudziko ebhizinesi.
Izvo zvinodikanwa kuti unzwisise kuti ketani yekugadzira inotanga nguva refu isati yagadzirwa modhi. Danho rayo rokutanga nderokutsanangura dambudziko rebhizimisi, fungidziro pamusoro pekukosha kunogona kutorwa kubva kune data, uye pfungwa yebhizimisi yekuishandisa.
Iwo chaiwo pfungwa yeMLOps yakasimuka seanofananidzira kune iyo pfungwa yeDevOps zvine chekuita nemhando yekudzidza yemuchina uye matekinoroji. DevOps inzira yekuvandudza software iyo inokutendera iwe kuti uwedzere kukurumidza kwekuita kwekuchinja kwemunhu uchichengetedza kuchinjika uye kuvimbika uchishandisa akati wandei nzira, kusanganisira kuenderera mberi, kupatsanura kwemabasa kuita akati wandei akazvimirira mamicroservices, otomatiki kuyedza uye kutumirwa kwemunhu. shanduko, kutarisa kwehutano hwepasirese, kukurumidza kupindura sisitimu yekutadza kunoonekwa, nezvimwe.
DevOps yakatsanangura hupenyu hwesoftware, uye nharaunda yauya nezano rekushandisa nzira imwechete kune data hombe. DataOps kuyedza kugadzirisa uye kuwedzera nzira uchifunga nezve maficha ekuchengetedza, kutumira uye kugadzirisa huwandu hukuru hwe data mumapuratifomu akasiyana uye anoshanda.
Nekuuya kwehumwe huremu hwakakosha hwemamodhi ekudzidza muchina akaiswa mumabhizinesi maitiro emabhizinesi, kufanana kwakasimba kwakaonekwa pakati pehupenyu hwemasvomhu ekudzidza muchina uye kutenderera kwesoftware. Musiyano chete ndewekuti modhi algorithms inogadzirwa uchishandisa muchina kudzidza maturusi uye nzira. Naizvozvo, iyo pfungwa yakasimuka kuti ishandise uye kugadzirisa nzira dzagara dzichizivikanwa pakuvandudza software yemamodhi ekudzidza muchina. Nekudaro, iwo anotevera matanho akakosha anogona kusiyaniswa mukutenderera kwehupenyu hwemashini ekudzidza modhi:
- kutsanangura pfungwa yebhizimisi;
- kudzidzisa kwemuenzaniso;
- kuedza uye kushandiswa kwemuenzaniso mukushanda kwebhizimisi;
- kushanda kwemuenzaniso.
Kana panguva yekushanda pane chido chekushandura kana kudzorerazve modhi pane data itsva, kutenderera kunotanga zvakare - muenzaniso unonatswa, wakaedzwa, uye shanduro itsva inoshandiswa.
Retreat. Sei uchidzokorora uye usingadzidzire zvakare? Izwi rokuti "muenzaniso wekudzidzira" rine zvarinoreva kaviri: pakati penyanzvi zvinoreva kukanganisa kwemuenzaniso, apo muenzaniso unofanotaura zvakanaka, unodzokorora kudzokorora parameter yakafanotaurwa pamusoro pegadziriro yekudzidzira, asi inoita zvakanyanya kuipa pane sampuli ye data yekunze. Nomuzvarirwo, muenzaniso wakadaro urema, sezvo chirema ichi hachibvumiri kushandiswa kwayo.
Mukutenderera kwehupenyu uku, zvinoita sekunge zvine musoro kushandisa maturusi eDevOps: otomatiki kuyedza, kutumira uye kutarisa, kugadzira maverengero emhando nenzira yeakaparadzana mamicroservices. Asi kune zvakare huwandu hwezvinhu zvinodzivirira kushandiswa kwakananga kwezvishandiso izvi pasina kuwedzera ML kusunga.
Maitiro ekuita kuti mamodheru ashande uye awane pundutso
Semuenzaniso waticharatidza kushandiswa kweMLOps maitiro, isu tichatora iro basa rekutanga rerobhoti yerutsigiro yekubhengi (kana chero chimwe) chigadzirwa. Kazhinji, bhizinesi rekutsigira bhizinesi rinotaridzika seizvi: mutengi anopinda meseji aine mubvunzo muhurukuro uye anogamuchira mhinduro kubva kune nyanzvi mukati meiyo yakafanotaurwa yenhaurirano muti. Basa rekuita otomatiki chat yakadaro rinowanzogadziriswa uchishandisa nyanzvi yakatsanangurwa seti yemitemo, iyo inonyanya kushanda nesimba kusimudzira nekuchengetedza. Kubudirira kweautomation yakadaro, zvichienderana nehuwandu hwekuoma kwebasa, inogona kuva 20-30%. Nomuzvarirwo, pfungwa inomuka kuti inobatsira zvakanyanya kushandisa artificial intelligence module - modhi yakagadzirwa uchishandisa muchina kudzidza, iyo:
- inokwanisa kugadzirisa nhamba huru yezvikumbiro pasina kubatanidzwa kwevashandi (zvichienderana nehurukuro, mune dzimwe nguva kubudirira kunogona kusvika 70-80%);
- inogadzirisa zvirinani kune isiri-yakajairwa mazwi munhaurirano - inokwanisa kuona chinangwa, chishuwo chaicho chemushandisi zvichienderana nechikumbiro chisina kunyatsogadzirwa;
- anoziva nzira yekuziva kana mhinduro yemuenzaniso yakakwana, uye kana pane kusava nechokwadi pamusoro pe "kuziva" kwemhinduro iyi uye unoda kubvunza mumwe mubvunzo wekujekesa kana kushandura kumushandisi;
- inogona kuwedzera kudzidziswa otomatiki (panzvimbo yeboka revagadziri vanogara vachigadzirisa uye kugadzirisa zvinyorwa zvekupindura, modhi yacho inodzidziswazve nenyanzvi yeData Sayenzi vachishandisa maraibhurari ekudzidza muchina akakodzera).
Nzira yekuita sei muenzaniso wepamusoro wakadaro ushande?
Sekugadzirisa chero rimwe dambudziko, usati wagadzira module yakadai, zvinodikanwa kutsanangura maitiro ebhizinesi uye kutsanangura zviri pamutemo iro rakananga basa ratinogadzirisa tichishandisa nzira yekudzidza muchina. Panguva ino, maitiro ekushanda, akasarudzwa neacronym Ops, anotanga.
Danho rinotevera nderekuti Data Scientist, pamwe nekubatana neData Engineer, inotarisa kuwanikwa uye kukwana kwedata uye fungidziro yebhizinesi nezve kugona kweiyo bhizinesi zano, kugadzira prototype modhi uye kuyedza kushanda kwayo chaiko. Chete mushure mekusimbiswa nebhizinesi kunogona shanduko kubva pakugadzira modhi kuenda kuibatanidza mumasystem anoita bhizinesi chairo kutanga. Kupera-kusvika-kumagumo kuronga kwekuita, kunzwisisa kwakadzama padanho rega rega rekuti modhi ichashandiswa sei uye kuti ichaunza hupfumi hupi, ipfungwa yakakosha mumatanho ekuunza nzira dzeMLOps mune yekambani tekinoroji mamiriro.
Nekuvandudzwa kweAI tekinoroji, huwandu uye akasiyana matambudziko anogona kugadziriswa uchishandisa muchina kudzidza ari kuwedzera zvakanyanya. Imwe neimwe yakadai bhizinesi maitiro ndeyekuchengetedza kambani nekuda kwekuita otomatiki kwevashandi vakawanda (call centre, kutarisa uye kuronga magwaro, nezvimwewo), iko kuwedzera kwebhesi revatengi nekuwedzera mabasa matsva anoyevedza uye ari nyore. iri kuchengetedza mari nekuda kwekushandisa kwavo zvakanyanya uye kugoverazve zviwanikwa nezvimwe zvakawanda. Pakupedzisira, chero maitiro anotarisa pakugadzira kukosha uye, semugumisiro, anofanira kuunza humwe hupfumi hunoita. Pano zvakakosha kuumba zvakajeka bhizinesi zano uye kuverenga iyo inotarisirwa purofiti kubva mukuita iyo modhi mune yakazara kukosha kwekugadzira chimiro chekambani. Pane mamiriro ezvinhu apo kushandisa modhi hakuzvipembedzi, uye nguva inoshandiswa nenyanzvi dzekudzidzira muchina inodhura zvakanyanya kupfuura nzvimbo yebasa yemushandisi ari kuita basa iri. Ndicho chikonzero nei zvakakosha kuedza kuziva nyaya dzakadaro mumatanho ekutanga ekugadzira AI systems.
Nekuda kweizvozvo, modhi dzinotanga kuunza purofiti chete kana dambudziko rebhizinesi ragadzirwa nenzira kwayo muhurongwa hweMLOps, zvinokosheswa zvakaiswa, uye maitiro ekuunza modhi muhurongwa akagadzirwa mumatanho ekutanga ebudiriro.
Maitiro matsva - matambudziko matsva
Mhinduro yakazara kumubvunzo wakakosha webhizinesi nezve mashandisiro anoita maML modhi pakugadzirisa matambudziko, iyo nyaya yekuvimba muAI nderimwe rematambudziko akakosha mukugadzira uye kuita nzira dzeMLOps. Pakutanga, mabhizinesi haana chokwadi nezvekutangwa kwemichina yekudzidza mumatanho - zvakaoma kuvimba nemhando munzvimbo dzaimbova, sekutonga, vanhu vaishanda. Kune bhizinesi, mapurogiramu anoita se "bhokisi dema", kukosha kwaro kuchiri kuda kuratidzwa. Mukuwedzera, mumabhengi, mubhizinesi revanofambisa telecom uye vamwe, pane zvakasimba zvinodiwa zvevatongi vehurumende. Ese masisitimu uye algorithms anoitwa mumabhangi maitiro ari pasi pekuongororwa. Kugadzirisa dambudziko iri, kuratidza kune bhizinesi uye regulators kutendeseka uye kurongeka kwemhinduro dzehungwaru hwekugadzira, maturusi ekutarisa ari kuunzwa pamwe nemuenzaniso. Uye zvakare, kune yakazvimirira nzira yekusimbisa, inosungirwa kune ekudzora modhi, inosangana nezvinodiwa neCentral Bank. Boka rakazvimirira renyanzvi rinoongorora zvawanikwa nemuenzaniso richifunga nezve data rekuisa.
Dambudziko rechipiri nderekuongorora uye kufunga nezve njodzi dzemodhi pakuita modhi yekudzidza yemuchina. Kunyangwe kana munhu asingakwanisi kupindura mubvunzo aine zana muzana rechokwadi kana chipfeko chimwe chete ichocho chaive chichena kana chebhuruu, saka njere dzekugadzira dzinewo kodzero yekukanganisa. Izvo zvakakoshawo kufunga kuti data rinogona kuchinja nekufamba kwenguva, uye modhi dzinoda kudzidziswazve kuitira kuti ibudise mhedzisiro yakakwana. Kuti ive nechokwadi chekuti bhizinesi haritambure, zvinodikanwa kubata njodzi dzemhando uye kutarisa mashandiro emuenzaniso, kugara uchiidzokorora pane data nyowani.
Asi mushure mechikamu chekutanga chekusavimbana, mhedzisiro inotanga kuoneka. Iwo mamodheru akawanda akaiswa zvinobudirira kuita maitiro, ndipo pakawedzera shungu dzebhizinesi rekushandisa njere dzekugadzira - matambudziko matsva uye matsva ari kuwanikwa anogona kugadziriswa uchishandisa nzira dzekudzidza dzemuchina. Basa rega rega rinokonzera hurongwa hwese hunoda humwe hunyanzvi:
- mainjiniya edata anogadzirira uye anogadzirisa data;
- data masayendisiti anoshandisa michina yekudzidza maturusi uye kugadzira modhi;
- IT inoshandisa iyo modhi muhurongwa;
- Injiniya yeML inosarudza nzira yekubatanidza iyi modhi nenzira kwayo, iyo maturusi eIT ekushandisa, zvichienderana nezvinodiwa zvemashandisirwo eiyo modhi, tichifunga nezvekuyerera kwezvikumbiro, nguva yekupindura, nezvimwe.
- Mugadziri weML anoronga kuti chigadzirwa chesoftware chinogona kuitwa sei muhurongwa hweindasitiri.
Kutenderera kwese kunoda nhamba huru yenyanzvi dzakanyatso hunyanzvi. Pane imwe nguva mukuvandudza uye dhigirii rekupinda kweML modhi mumabhizinesi maitiro, zvinoitika kuti mutsara kuyera huwandu hwenyanzvi maererano nekuwedzera kwehuwandu hwemabasa kunodhura uye kusashanda. Naizvozvo, mubvunzo unomuka wekuita otomatiki iyo MLOps maitiro - kutsanangura akati wandei akajairwa makirasi ematambudziko ekudzidza muchina, kugadzira akajairwa data kugadzirisa mapaipi uye kumwe kudzidziswa kwemamodhi. Mumufananidzo wakakodzera, kugadzirisa matambudziko akadai kunoda nyanzvi dzine hunyanzvi muhunyanzvi pamharadzano yeBig Data, Data Science, DevOps uye IT. Naizvozvo, dambudziko hombe muindasitiri yeData Science uye dambudziko rakakura mukuronga MLOps maitirwo ndiko kushomeka kwehunyanzvi hwakadaro mumusika uripo wekudzidzisa. Nyanzvi dzinosangana nezvinodiwa izvi parizvino hadziwanzo pamusika wevashandi uye dzinokosha uremu hwavo mugoridhe.
Panyaya yehunyanzvi
Mune dzidziso, ese MLOps mabasa anogona kugadziriswa uchishandisa yekare DevOps maturusi uye pasina kushandisa kune yakasarudzika yekuwedzera yemuenzaniso. Zvino, sezvatakaona pamusoro, sainzi wedata haafanire kunge ari nyanzvi yemasvomhu uye muongorori wedata, asiwo guru repombi yese - ane basa rekugadzira zvivakwa, madhizaini emhando mumitauro yakati wandei zvichienderana nekuvakwa, kugadzirira. a data mart uye kuendesa iyo application pachayo. Nekudaro, kugadzira hunyanzvi hwetekinoroji hunoitwa mukupedzisira-kusvika-kumagumo MLOps maitiro kunotora kusvika 80% yemitengo yevashandi, zvinoreva kuti nyanzvi yemasvomhu inokwanisa, uyo ari wemhando yeData Scientist, achapa chete makumi maviri muzana yenguva yake kuhunyanzvi hwake. . Naizvozvo, kutsanangura mabasa enyanzvi anobatanidzwa mukuita kwekushandisa michina yekudzidza modhi inova yakakosha.
Kuti madonzo anofanirwa kutsanangurwa sei zvinoenderana nehukuru hwebhizinesi. Chinhu chimwe chete kana yekutanga iine nyanzvi imwe chete, anoshanda nesimba munzvimbo yekuchengetedza simba, anova mainjiniya wake, mugadziri wezvivakwa, uye DevOps. Iyo inyaya yakasiyana zvachose apo, mune bhizinesi hombe, ese emhando yekusimudzira maitiro akaiswa pane mashoma epamusoro-danho reData Science nyanzvi, nepo programmer kana dhatabhesi nyanzvi - yakajairika uye isingadhure kugona mumusika wevashandi - inogona kutora. pazvizhinji zvebasa.mabasa enguva dzose.
Saka, kumhanya uye mhando yemhando dzakagadzirwa, kugadzirwa kwechikwata uye microclimate mairi zvakananga zvinoenderana nekuti muganho uri papi mukusarudzwa kwenyanzvi kutsigira chirongwa cheMLOps uye kuti maitiro ekushandiswa kwemhando dzakagadzirwa akarongwa sei. .
Zvatoitwa nechikwata chedu
Isu nguva pfupi yadarika takatanga kuvaka chimiro chehunyanzvi uye maMLOps maitiro. Asi mapurojekiti edu emumodhi yehupenyu manejimendi uye pakushandisa modhi sevhisi atove mudanho rekuyedza MVP.
Isu takasarudzawo iyo yakakwana yekugona chimiro chebhizinesi hombe uye chimiro chesangano chekudyidzana pakati pevatori vechikamu mukuita. Mapoka eAgile akarongwa kuti agadzirise matambudziko emhando yese yevatengi vebhizinesi, uye nzira yekudyidzana nemapoka epurojekiti kugadzira mapuratifomu uye zvivakwa, iyo ndiyo nheyo yeMLOps chivako chiri kuvakwa, yakasimbiswa.
Mibvunzo yeramangwana
MLOps inzvimbo iri kukura iri kusangana nekushomeka kwehunyanzvi uye ichawedzera simba mune ramangwana. Panguva ino, zviri nani kuvaka paDevOps kuvandudza uye maitiro. Chinangwa chikuru cheMLOps ndechekushandisa zvakanyanya ML modhi kugadzirisa matambudziko ebhizinesi. Asi izvi zvinomutsa mibvunzo yakawanda:
- Nzira yekudzikisa sei nguva yekutanga mamodheru mukugadzira?
- Nzira yekudzikisa sei kukakavara pakati pezvikwata zvehunyanzvi hwakasiyana uye kuwedzera kutarisa kwekubatana?
- Maitiro ekutevera mamodheru, maneja mavhezheni uye kuronga kunoshanda kutarisa?
- Maitiro ekugadzira yechokwadi denderedzwa lifecycle yemazuva ano ML modhi?
- Nzira yekumisa sei muchina kudzidza maitiro?
Mhinduro dzemibvunzo iyi dzichanyanya kuona kuti MLOps inosvika sei pakukwanisa kwayo kuzere.
Source: www.habr.com