{"id":86523,"date":"2020-06-26T07:42:14","date_gmt":"2020-06-26T05:42:14","guid":{"rendered":"https:\/\/prohoster.info\/blog\/administrirovanie\/mlops-devops-v-mire-machine-learning"},"modified":"2020-06-26T07:42:14","modified_gmt":"2020-06-26T05:42:14","slug":"mlops-devops-v-mire-machine-learning","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sk\/blog\/administrirovanie\/mlops-devops-v-mire-machine-learning","title":{"rendered":"MLOps: DevOps vo svete strojov\u00e9ho u\u010denia","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"<p>V roku 2018 sa v profesion\u00e1lnych kruhoch a na tematick\u00fdch konferenci\u00e1ch venovan\u00fdch AI objavil pojem MLOps, ktor\u00fd sa r\u00fdchlo upevnil v odvetv\u00ed a teraz sa vyv\u00edja ako samostatn\u00fd smer. V bud\u00facnosti m\u00f4\u017ee MLOps sta\u0165 jednou z naj\u017eiadanej\u0161\u00edch oblast\u00ed v IT. \u010co to vlastne je a ako to funguje, dozvieme sa \u010falej.<\/p>\n<p><img decoding=\"async\" alt=\"MLOps: DevOps vo svete strojov\u00e9ho u\u010denia\" src=\"\/wp-content\/uploads\/2020\/06\/03bcaaaf5b9ddd864c7e8e0a607d81c4.jpeg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<noindex><a rel=\"nofollow\" name=\"habracut\"><\/a><\/noindex><\/p>\n<h2>\u010co je MLOps<\/h2>\n<p>\nMLOps (spojenie technol\u00f3gi\u00ed a procesov strojov\u00e9ho u\u010denia a pr\u00edstupov k implement\u00e1cii vyvinut\u00fdch modelov do obchodn\u00fdch procesov) je nov\u00fd sp\u00f4sob spolupr\u00e1ce medzi z\u00e1stupcami biznisu, vedcami, matematikmi, \u0161pecialistami v oblasti strojov\u00e9ho u\u010denia a IT in\u017einiermi pri vytv\u00e1ran\u00ed syst\u00e9mov umelej inteligencie.<\/p>\n<p>In\u00fdmi slovami, je to sp\u00f4sob, ako premeni\u0165 met\u00f3dy a technol\u00f3gie strojov\u00e9ho u\u010denia na u\u017eito\u010dn\u00fd n\u00e1stroj pre rie\u0161enie obchodn\u00fdch \u00faloh.\u00a0<\/p>\n<p>Je potrebn\u00e9 ch\u00e1pa\u0165, \u017ee re\u0165azec produktiz\u00e1cie za\u010d\u00edna d\u00e1vno pred v\u00fdvojom modelu. Jeho prv\u00fdm krokom je definova\u0165 obchodn\u00fd probl\u00e9m, hypot\u00e9zy o hodnote, ktor\u00fa je mo\u017en\u00e9 z d\u00e1t vy\u0165a\u017ei\u0165, a obchodn\u00e9 my\u0161lienky na jej aplik\u00e1ciu.\u00a0<\/p>\n<p>Pojem MLOps vznikol ako anal\u00f3gia pojmu DevOps aplikovan\u00e1 na modely a technol\u00f3gie strojov\u00e9ho u\u010denia. DevOps je pr\u00edstup k v\u00fdvoju softv\u00e9ru, ktor\u00fd umo\u017e\u0148uje zv\u00fd\u0161i\u0165 r\u00fdchlos\u0165 nasadzovania jednotliv\u00fdch zmien pri zachovan\u00ed flexibility a spo\u013eahlivosti pomocou r\u00f4znych pr\u00edstupov, medzi ktor\u00e9 patr\u00ed nepretr\u017eit\u00fd v\u00fdvoj, rozdelenie funkci\u00ed na radu nez\u00e1visl\u00fdch mikroservisov, automatizovan\u00e9 testovanie a nasadenie jednotliv\u00fdch zmien, glob\u00e1lny monitoring funk\u010dnosti, syst\u00e9m r\u00fdchlej reakcie na zisten\u00e9 poruchy a \u010fal\u0161ie.\u00a0<\/p>\n<p>DevOps definoval \u017eivotn\u00fd cyklus softv\u00e9ru a v komunite odborn\u00edkov sa objavila my\u0161lienka pou\u017ei\u0165 t\u00fato met\u00f3du aj na ve\u013ek\u00e9 d\u00e1ta. DataOps je pokus prisp\u00f4sobi\u0165 a roz\u0161\u00edri\u0165 t\u00fato met\u00f3du s oh\u013eadom na \u0161pecifik\u00e1 ukladania, prenosu a spracovania ve\u013ek\u00fdch objemov d\u00e1t na r\u00f4znych a vz\u00e1jomne si rozumiacich platform\u00e1ch.<br \/>\n\u00a0\u00a0<br \/>\nS pr\u00edchodom ur\u010ditej kritickej masy modelov strojov\u00e9ho u\u010denia integrovan\u00fdch do podnikov\u00fdch procesov sa pozorovalo siln\u00e9 podobn\u00e9 spr\u00e1vanie \u017eivotn\u00e9ho cyklu matematick\u00fdch modelov strojov\u00e9ho u\u010denia a \u017eivotn\u00e9ho cyklu softv\u00e9ru. Jedin\u00fd rozdiel je v tom, \u017ee algoritmy modelov sa vytv\u00e1raj\u00fa pomocou n\u00e1strojov a met\u00f3d strojov\u00e9ho u\u010denia. Preto prirodzene vznikla my\u0161lienka aplikova\u0165 a prisp\u00f4sobi\u0165 u\u017e zn\u00e1me pr\u00edstupy k v\u00fdvoju softv\u00e9ru pre modely strojov\u00e9ho u\u010denia. Tak\u017ee v \u017eivotnom cykle modelov strojov\u00e9ho u\u010denia m\u00f4\u017eeme vy\u010dleni\u0165 nasleduj\u00face k\u013e\u00fa\u010dov\u00e9 etapy:<\/p>\n<ul>\n<li>definovanie obchodnej my\u0161lienky;\n<\/li>\n<li>tr\u00e9novanie modelu;\n<\/li>\n<li>testovanie a implement\u00e1cia modelu do podnikov\u00e9ho procesu;\n<\/li>\n<li>prev\u00e1dzka modelu.\n<\/li>\n<\/ul>\n<p>\nKe\u010f v priebehu prev\u00e1dzky nastane potreba zmeni\u0165 alebo preu\u010di\u0165 model na nov\u00fdch d\u00e1tach, cyklus sa za\u010d\u00edna odznova \u2014 model sa vylep\u0161uje, testuje a nasadzuje sa nov\u00e1 verzia.<\/p>\n<blockquote><p>Odklon. Pre\u010do preu\u010di\u0165, a nie znovu preu\u010di\u0165? Term\u00edn \u201eznovu preu\u010denie modelu\u201c m\u00e1 dvojit\u00fd v\u00fdznam: medzi odborn\u00edkmi znamen\u00e1 chybu modelu, ke\u010f model dobre predpoved\u00e1, v skuto\u010dnosti opakuje predpovedan\u00fd parameter na tr\u00e9novacej vzorke, ale ove\u013ea hor\u0161ie funguje na extern\u00fdch d\u00e1tach. Samozrejme, tak\u00fdto model je pova\u017eovan\u00fd za zmetok, preto\u017ee t\u00e1to chyba br\u00e1ni jeho aplik\u00e1cii.<\/p><\/blockquote>\n<p>\nV tomto \u017eivotnom cykle sa logicky jav\u00ed pou\u017eitie n\u00e1strojov DevOps: automatizovan\u00e9 testovanie, nasadenie a monitorovanie, prezent\u00e1cia v\u00fdpo\u010dtov modelov vo forme samostatn\u00fdch mikroslu\u017eieb. Ale existuje aj nieko\u013eko \u0161pecifick\u00fdch aspektov, ktor\u00e9 br\u00e1nia priamemu pou\u017eitiu t\u00fdchto n\u00e1strojov bez dodato\u010dnej ML v\u00e4zby.<\/p>\n<p><img decoding=\"async\" alt=\"MLOps: DevOps vo svete strojov\u00e9ho u\u010denia\" src=\"\/wp-content\/uploads\/2020\/06\/e1d01fd68c3f0071f925a82c08d29d25.jpeg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<\/p>\n<h2>Ako prin\u00fati\u0165 modely pracova\u0165 a prin\u00e1\u0161a\u0165 zisk<\/h2>\n<p>\nAko pr\u00edklad, na ktorom predvedieme aplik\u00e1ciu pr\u00edstupu MLOps, sa zameriame na klasick\u00fa \u00falohu automatiz\u00e1cie chatu z\u00e1kazn\u00edckej podpory bankov\u00e9ho (alebo ak\u00e9hoko\u013evek in\u00e9ho) produktu. Zvy\u010dajne obchodn\u00fd proces podpory prostredn\u00edctvom chatu vyzer\u00e1 nasledovne: z\u00e1kazn\u00edk zad\u00e1 do chatu spr\u00e1vu s ot\u00e1zkou a dostane odpove\u010f od \u0161pecialistu v r\u00e1mci vopred definovanej \u0161trukt\u00fary dial\u00f3gu. \u00daloha automatiz\u00e1cie tak\u00e9hoto chatu sa obvykle rie\u0161i pomocou expertn\u00fdch s\u00faborov pravidiel, ktor\u00e9 s\u00fa ve\u013emi pr\u00e1cne na v\u00fdvoj a \u00fadr\u017ebu. Efekt\u00edvnos\u0165 takejto automatiz\u00e1cie, v z\u00e1vislosti od \u00farovne zlo\u017eitosti \u00falohy, m\u00f4\u017ee dosiahnu\u0165 20\u201330%. Samozrejme, vznik\u00e1 my\u0161lienka, \u017ee v\u00fdhodnej\u0161ie je zavedenie modulu umelej inteligencie \u2014 modelu vyvinut\u00e9ho pomocou strojov\u00e9ho u\u010denia, ktor\u00fd:<\/p>\n<ul>\n<li>je schopn\u00fd spracova\u0165 bez \u00fa\u010dasti oper\u00e1tora v\u00e4\u010d\u0161ie mno\u017estvo po\u017eiadaviek (v z\u00e1vislosti od t\u00e9my, v niektor\u00fdch pr\u00edpadoch m\u00f4\u017ee efektivita dosiahnu\u0165 70\u201380%);\n<\/li>\n<li>lep\u0161ie sa prisp\u00f4sobuje ne\u0161tandardn\u00fdm formul\u00e1ci\u00e1m v dial\u00f3gu \u2014 dok\u00e1\u017ee ur\u010di\u0165 intent, skuto\u010dn\u00e9 \u017eelanie pou\u017e\u00edvate\u013ea na z\u00e1klade nepresne formulovanej po\u017eiadavky;\n<\/li>\n<li>dok\u00e1\u017ee ur\u010di\u0165, kedy je odpove\u010f modelu adekv\u00e1tna, a kedy existuj\u00fa pochybnosti o \u201evedomosti\u201c tejto odpovede a je potrebn\u00e9 polo\u017ei\u0165 dodato\u010dn\u00fa spres\u0148uj\u00facu ot\u00e1zku alebo prep\u00edna\u0165 na oper\u00e1tora;\n<\/li>\n<li>m\u00f4\u017ee by\u0165 automatizovane \u010falej tr\u00e9novan\u00e1 (namiesto skupiny v\u00fdvoj\u00e1rov, ktor\u00ed neust\u00e1le prisp\u00f4sobuj\u00fa a upravuj\u00fa scen\u00e1re odpoved\u00ed, model \u010falej tr\u00e9nuje \u0161pecialista na Data Science, ktor\u00fd pou\u017e\u00edva pr\u00edslu\u0161n\u00e9 kni\u017enice strojov\u00e9ho u\u010denia).\u00a0\n<\/li>\n<\/ul>\n<p>\n<img decoding=\"async\" alt=\"MLOps: DevOps vo svete strojov\u00e9ho u\u010denia\" src=\"\/wp-content\/uploads\/2020\/06\/9ef6356090a250df6cb212ab8bedb5f3.jpeg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nAko prin\u00fati\u0165 tak\u00fd pokro\u010dil\u00fd model pracova\u0165?\u00a0<\/p>\n<p>Rovnako ako pri rie\u0161en\u00ed akejko\u013evek inej \u00falohy, pred t\u00fdm, ne\u017e vyvineme tak\u00fdto modul, je potrebn\u00e9 definova\u0165 obchodn\u00fd proces a form\u00e1lne op\u00edsa\u0165 konkr\u00e9tnu \u00falohu, ktor\u00fa budeme rie\u0161i\u0165 pomocou met\u00f3dy strojov\u00e9ho u\u010denia. V tomto bode za\u010d\u00edna proces operacionaliz\u00e1cie ozna\u010den\u00fd skratkou Ops.\u00a0<\/p>\n<p>Nasleduj\u00facim krokom \u0161pecialista na Data Science v spolupr\u00e1ci s d\u00e1tov\u00fdm in\u017einierom kontroluje dostupnos\u0165 a dostato\u010dnos\u0165 d\u00e1t a obchodn\u00fa hypot\u00e9zu o funk\u010dnosti obchodnej my\u0161lienky, vytv\u00e1ra prototyp modelu a testuje jeho skuto\u010dn\u00fa efekt\u00edvnos\u0165. A\u017e po potvrden\u00ed zo strany podnikania m\u00f4\u017ee za\u010da\u0165 prechod od v\u00fdvoja modelu k jeho implement\u00e1cii do syst\u00e9mov vykon\u00e1vaj\u00facich konkr\u00e9tne obchodn\u00e9 procesy. Prie\u010dne pl\u00e1novanie implement\u00e1cie, hlbok\u00e9 pochopenie v ka\u017edej f\u00e1ze, ako bude model pou\u017e\u00edvan\u00fd a ak\u00fd ekonomick\u00fd pr\u00ednos prinesie, je z\u00e1kladn\u00fdm bodom v procesoch implement\u00e1cie pr\u00edstupov MLOps do technologick\u00e9ho prostredia spolo\u010dnosti.<\/p>\n<p>S rozvojom technol\u00f3gi\u00ed AI narast\u00e1 mno\u017estvo a rozmanitos\u0165 \u00faloh, ktor\u00e9 je mo\u017en\u00e9 rie\u0161i\u0165 pomocou strojov\u00e9ho u\u010denia. Ka\u017ed\u00fd tak\u00fdto obchodn\u00fd proces znamen\u00e1 \u00fasporu pre spolo\u010dnos\u0165 v\u010faka automatiz\u00e1cii pr\u00e1ce zamestnancov v masov\u00fdch poz\u00edci\u00e1ch (call centrum, kontrola a triedenie dokumentov at\u010f.), roz\u0161\u00edrenie z\u00e1kazn\u00edckej z\u00e1kladne pridan\u00edm nov\u00fdch atrakt\u00edvnych a pohodln\u00fdch funkci\u00ed, \u00fasporu prostriedkov optimaliz\u00e1ciou ich vyu\u017eitia a redistrib\u00faciou zdrojov a mnoho \u010fal\u0161ieho. Napokon, ka\u017ed\u00fd proces je zameran\u00fd na vytv\u00e1ranie hodnoty a ako d\u00f4sledok by mal prin\u00e1\u0161a\u0165 konkr\u00e9tny ekonomick\u00fd efekt. Je ve\u013emi d\u00f4le\u017eit\u00e9 jasne formulate obchodn\u00fa ideu a vypo\u010d\u00edta\u0165 predpokladan\u00fd zisk z implement\u00e1cie modelu v celkovej \u0161trukt\u00fare vytv\u00e1rania hodnoty spolo\u010dnosti. Niekedy nast\u00e1vaj\u00fa situ\u00e1cie, ke\u010f implement\u00e1cia modelu sa neosved\u010d\u00ed, a \u010das, ktor\u00fd odborn\u00edci na strojov\u00e9 u\u010denie str\u00e1via, sa uk\u00e1\u017ee by\u0165 ove\u013ea drah\u0161\u00ed ako pracovn\u00e9 miesto oper\u00e1tora, ktor\u00fd vykon\u00e1va t\u00fato \u00falohu. Preto je potrebn\u00e9 tak\u00e9to pr\u00edpady sna\u017ei\u0165 sa odhali\u0165 v ran\u00fdch f\u00e1zach vytv\u00e1rania syst\u00e9mov AI.<\/p>\n<p>Z toho vypl\u00fdva, \u017ee zisk modelu za\u010d\u00edna prich\u00e1dza\u0165 iba vtedy, ke\u010f bola v procese MLOps spr\u00e1vne formulovan\u00e1 obchodn\u00e1 \u00faloha, ur\u010den\u00e9 priority a v ran\u00fdch f\u00e1zach v\u00fdvoja bol definovan\u00fd proces implement\u00e1cie modelu do syst\u00e9mu.<\/p>\n<h2>Nov\u00fd proces \u2013 nov\u00e9 v\u00fdzvy<\/h2>\n<p>\nVy\u010derp\u00e1vaj\u00faca odpove\u010f na z\u00e1sadn\u00fa ot\u00e1zku podnikania, do akej miery s\u00fa modely strojov\u00e9ho u\u010denia aplikovate\u013en\u00e9 na rie\u0161enie probl\u00e9mov, a v\u0161eobecn\u00e1 ot\u00e1zka d\u00f4very v AI \u2013 to je jedna z k\u013e\u00fa\u010dov\u00fdch v\u00fdziev v procese rozvoja a implement\u00e1cie pr\u00edstupov MLOps. Podniky spo\u010diatku skepticky pristupuj\u00fa k implement\u00e1cii strojov\u00e9ho u\u010denia do procesov \u2013 \u0165a\u017eko sa spolieha\u0165 na modely v oblastiach, kde tradi\u010dne pracovali \u013eudia. Pre podniky sa programy javia ako \u201e\u010dierna skrinka\u201c, ktorej relevantnos\u0165 odpoved\u00ed sa e\u0161te mus\u00ed dok\u00e1za\u0165. Okrem toho v bankovom sektore, v oblasti telekomunika\u010dn\u00fdch oper\u00e1torov a v in\u00fdch oblastiach existuj\u00fa pr\u00edsne po\u017eiadavky vl\u00e1dnych regul\u00e1torov. V\u0161etky syst\u00e9my a algoritmy implementovan\u00e9 v bankov\u00fdch procesoch podliehaj\u00fa auditu. Aby sa t\u00e1to \u00faloha vyrie\u0161ila a dok\u00e1zala sa podnikom a regul\u00e1torom opr\u00e1vnenos\u0165 a spr\u00e1vnos\u0165 odpoved\u00ed umelej inteligencie, s\u00fa spolu s modelom implementovan\u00e9 monitorovacie n\u00e1stroje. Existuje aj proces nez\u00e1vislej valid\u00e1cie, ktor\u00fd je povinn\u00fd pre regula\u010dn\u00e9 modely, ktor\u00fd sp\u013a\u0148a po\u017eiadavky centr\u00e1lnej banky. Nez\u00e1visl\u00e1 expert\u00edza vykon\u00e1va audit v\u00fdsledkov, ktor\u00e9 model z\u00edskal, pri\u010dom zoh\u013ead\u0148uje vstupn\u00e9 \u00fadaje.<\/p>\n<p>Druh\u00e1 v\u00fdzva \u2013 hodnotenie a zoh\u013eadnenie riz\u00edk modelov pri implement\u00e1cii modelu strojov\u00e9ho u\u010denia. Ak ani \u010dlovek nem\u00f4\u017ee so stopercentnou istotou odpoveda\u0165 na ot\u00e1zku, \u010di bolo to \u0161aty biele alebo modr\u00e9, tak aj umel\u00e1 inteligencia m\u00e1 pr\u00e1vo na chybu. Je tie\u017e potrebn\u00e9 zoh\u013eadni\u0165, \u017ee \u010dasom sa \u00fadaje m\u00f4\u017eu meni\u0165 a modely sa musia \u010falej u\u010di\u0165, aby poskytovali dostato\u010dne presn\u00fd v\u00fdsledok. Aby biznis proces netrpel, je potrebn\u00e9 riadi\u0165 rizik\u00e1 modelov a sledova\u0165 pr\u00e1cu modelu, pravidelne ho pre\u0161ko\u013eova\u0165 na nov\u00fdch \u00fadajoch.<\/p>\n<p><img decoding=\"async\" alt=\"MLOps: DevOps vo svete strojov\u00e9ho u\u010denia\" src=\"\/wp-content\/uploads\/2020\/06\/cb97ff0eab95d81e17bf76aad3c3994e.jpeg\" style=\"display:block;margin: 0 auto;\" \/><br \/>\n<br \/>\nAle po prvej f\u00e1ze ned\u00f4very sa za\u010d\u00edna prejavova\u0165 opa\u010dn\u00fd efekt. \u010c\u00edm viac modelov sa \u00faspe\u0161ne implementuje do procesov, t\u00fdm viac rastie apet\u00edt podnikov po vyu\u017eit\u00ed umelej inteligencie \u2013 nach\u00e1dzaj\u00fa sa nov\u00e9 a nov\u00e9 \u00falohy, ktor\u00e9 je mo\u017en\u00e9 rie\u0161i\u0165 met\u00f3dami strojov\u00e9ho u\u010denia. Ka\u017ed\u00e1 \u00faloha sp\u00fa\u0161\u0165a cel\u00fd proces, ktor\u00fd si vy\u017eaduje r\u00f4zne kompetencie:<\/p>\n<ul>\n<li>d\u00e1tov\u00ed in\u017einieri pripravuj\u00fa a spracov\u00e1vaj\u00fa \u00fadaje;\n<\/li>\n<li>d\u00e1tov\u00ed vedci aplicuj\u00fa n\u00e1stroje strojov\u00e9ho u\u010denia a vyv\u00edjaj\u00fa model;\n<\/li>\n<li>IT zav\u00e1dza model do syst\u00e9mu;\n<\/li>\n<li>ML in\u017einier ur\u010duje, ako t\u00fato model spr\u00e1vne integrova\u0165 do procesu, ak\u00e9 IT n\u00e1stroje pou\u017ei\u0165 v z\u00e1vislosti od po\u017eiadaviek na re\u017eim aplik\u00e1cie modelu s oh\u013eadom na tok \u017eiadost\u00ed, \u010das odpovede a podobne.\u00a0\n<\/li>\n<li>ML architekt navrhuje, ako fyzicky realizova\u0165 softv\u00e9rov\u00fd produkt v priemyselnom syst\u00e9me.\n<\/li>\n<\/ul>\n<p>\nCel\u00fd cyklus vy\u017eaduje mno\u017estvo vysoko kvalifikovan\u00fdch odborn\u00edkov. V ur\u010ditom bode rozvoja a stup\u0148a prenikania ML modelov do obchodn\u00fdch procesov sa uk\u00e1\u017ee, \u017ee line\u00e1rne \u0161k\u00e1lova\u0165 po\u010det odborn\u00edkov \u00famerne k rastu po\u010dtu \u00faloh sa st\u00e1va n\u00e1kladn\u00e9 a neefekt\u00edvne. Preto vznik\u00e1 ot\u00e1zka automatiz\u00e1cie procesu MLOps \u2014 defin\u00edcia nieko\u013ek\u00fdch \u0161tandardn\u00fdch tried \u00faloh strojov\u00e9ho u\u010denia, v\u00fdvoj typov\u00fdch pipeline spracovania d\u00e1t a do\u0161ko\u013eovania modelov. V ide\u00e1lnom pr\u00edpade na rie\u0161enie tak\u00fdchto \u00faloh potrebuj\u00fa odborn\u00edci, ktor\u00ed rovnako dobre ovl\u00e1daj\u00fa kompetencie na rozhran\u00ed BigData, Data Science, DevOps a IT. Preto je najv\u00e4\u010d\u0161\u00edm probl\u00e9mom v priemysle Data Science a najv\u00e4\u010d\u0161ou v\u00fdzvou pri organiz\u00e1cii procesov MLOps \u2014 nedostatok takejto kompetencie na existuj\u00facom trhu pr\u00edpravy pracovnej sily. Odborn\u00edci, ktor\u00ed sp\u013a\u0148aj\u00fa tak\u00e9to po\u017eiadavky, s\u00fa v s\u00fa\u010dasnosti jedin\u00ed na trhu pr\u00e1ce a s\u00fa cenen\u00ed na v\u00e1hu zlata.<\/p>\n<h2>K ot\u00e1zke kompetenci\u00ed<\/h2>\n<p>\nTeoreticky sa v\u0161etky \u00falohy MLOps daj\u00fa rie\u0161i\u0165 klasick\u00fdmi n\u00e1strojmi DevOps a nemus\u00ed sa uch\u00fdli\u0165 k \u0161pecializovan\u00e9mu roz\u0161\u00edreniu rolnej model. Ako sme u\u017e povedali vy\u0161\u0161ie, d\u00e1tov\u00fd vedec by mal by\u0165 nielen matematikom a odborn\u00edkom na analytiku d\u00e1t, ale aj guru v\u0161etk\u00fdch pipeline \u2014 na jeho pleciach spo\u010d\u00edva v\u00fdvoj architekt\u00fary, programovanie modelov v nieko\u013ek\u00fdch jazykoch v z\u00e1vislosti od architekt\u00fary, pr\u00edprava vitr\u00edny d\u00e1t a nasadenie samotnej aplik\u00e1cie. Av\u0161ak vytvorenie technologickej ob\u00e1lky, realizovanej v procesoch MLOps, zaber\u00e1 a\u017e 80% pracovn\u00e9ho \u010dasu, \u010do znamen\u00e1, \u017ee kvalifikovan\u00fd matematik, ktor\u00fdm je kvalitn\u00fd d\u00e1tov\u00fd vedec, bude len 20% \u010dasu venova\u0165 svojej \u0161pecializ\u00e1cii. Preto sa rozdelenie rol\u00ed \u0161pecialistov, ktor\u00ed realizuj\u00fa proces implement\u00e1cie modelov strojov\u00e9ho u\u010denia, st\u00e1va \u017eivotne d\u00f4le\u017eit\u00fdm.\u00a0<\/p>\n<p>To what extent roles should be delineated depends on the scale of the enterprise. It's one thing when a startup has one specialist, a jack-of-all-trades who acts as an engineer, architect, and DevOps all at once. It's quite another when a large company centralizes all model development processes among a few high-level Data Science specialists, while programmers or database specialists\u2014a more common and less costly competence in the job market\u2014can take on most routine tasks.<\/p>\n<p>Thus, the boundary in selecting specialists to ensure the MLOps process and how the operationalization of developed models is organized directly impacts the speed and quality of model development, the productivity of the team, and the microclimate within it.<\/p>\n<h2>What our team has already accomplished<\/h2>\n<p>\nNot long ago, we began building the competency structure and MLOps processes. However, we currently have our projects on model lifecycle management and model deployment as a service in the MVP testing stage.<\/p>\n<p>We have also defined the optimal competency structure and organizational interaction for a large enterprise. Agile teams have been organized to address tasks for the entire spectrum of business clients, and a process for collaboration with project teams in creating platforms and infrastructure, which form the foundation of the MLOps framework, has been established.<\/p>\n<h2>Questions for the future<\/h2>\n<p>\nMLOps is an emerging field that is facing a shortage of competencies and will gain momentum in the future. For now, it is best to rely on the developments and practices of DevOps. The primary goal of MLOps is to utilize ML models more effectively to solve business problems. However, this raises many questions:<\/p>\n<ul>\n<li>How can we reduce the time to launch models into production?\n<\/li>\n<li>How can we minimize bureaucratic friction between teams of different competencies and enhance collaborative focus?\n<\/li>\n<li>How can we track models, manage versions, and organize effective monitoring?\n<\/li>\n<li>How can we create a truly cyclical lifecycle for a modern ML model?\n<\/li>\n<li>How can we standardize the machine learning process?\n<\/li>\n<\/ul>\n<p>\nOdpovede na tieto ot\u00e1zky do ve\u013ekej miery ur\u010dia, ako r\u00fdchlo MLOps odhal\u00ed svoj pln\u00fd potenci\u00e1l.<br \/>\n<br \/>Zdroj: <a content=\"nofollow\" rel=\"nofollow\" href=\"https:\/\/habr.com\/ru\/company\/vtb\/blog\/508012\/\">habr.com<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"<p>\u0412 2018 \u0433\u043e\u0434\u0443 \u0432 \u043f\u0440\u043e\u0444\u0435\u0441\u0441\u0438\u043e\u043d\u0430\u043b\u044c\u043d\u044b\u0445 \u043a\u0440\u0443\u0433\u0430\u0445 \u0438 \u043d\u0430 \u0442\u0435\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0445 \u043a\u043e\u043d\u0444\u0435\u0440\u0435\u043d\u0446\u0438\u044f\u0445, \u043f\u043e\u0441\u0432\u044f\u0449\u0435\u043d\u043d\u044b\u0445 AI, \u043f\u043e\u044f\u0432\u0438\u043b\u043e\u0441\u044c \u043f\u043e\u043d\u044f\u0442\u0438\u0435 MLOps, \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u0431\u044b\u0441\u0442\u0440\u043e \u0437\u0430\u043a\u0440\u0435\u043f\u0438\u043b\u043e\u0441\u044c \u0432 \u043e\u0442\u0440\u0430\u0441\u043b\u0438 \u0438 \u0441\u0435\u0439\u0447\u0430\u0441 \u0440\u0430\u0437\u0432\u0438\u0432\u0430\u0435\u0442\u0441\u044f \u043a\u0430\u043a \u0441\u0430\u043c\u043e\u0441\u0442\u043e\u044f\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043d\u0430\u043f\u0440\u0430\u0432\u043b\u0435\u043d\u0438\u0435. \u0412 \u043f\u0435\u0440\u0441\u043f\u0435\u043a\u0442\u0438\u0432\u0435 MLOps \u043c\u043e\u0436\u0435\u0442 \u0441\u0442\u0430\u0442\u044c \u043e\u0434\u043d\u043e\u0439 \u0438\u0437 \u043d\u0430\u0438\u0431\u043e\u043b\u0435\u0435 \u0432\u043e\u0441\u0442\u0440\u0435\u0431\u043e\u0432\u0430\u043d\u043d\u044b\u0445 \u0441\u0444\u0435\u0440 \u0432 IT. \u0427\u0442\u043e \u0436\u0435 \u044d\u0442\u043e \u0442\u0430\u043a\u043e\u0435 \u0438 \u0441 \u0447\u0435\u043c \u0435\u0433\u043e \u0435\u0434\u044f\u0442, \u0440\u0430\u0437\u0431\u0438\u0440\u0430\u0435\u043c\u0441\u044f \u043f\u043e\u0434 \u043a\u0430\u0442\u043e\u043c. \u0427\u0442\u043e \u0442\u0430\u043a\u043e\u0435 MLOps MLOps (\u043e\u0431\u044a\u0435\u0434\u0438\u043d\u0435\u043d\u0438\u0435 [&hellip;]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":1,"featured_media":86524,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[688],"tags":[],"class_list":["post-86523","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-administrirovanie"],"aioseo_notices":[],"aioseo_head":"\n\t\t<!-- All in One SEO 5.0.5 - aioseo.com -->\n\t<meta name=\"description\" content=\"\u0412 2018 \u0433\u043e\u0434\u0443 \u0432 \u043f\u0440\u043e\u0444\u0435\u0441\u0441\u0438\u043e\u043d\u0430\u043b\u044c\u043d\u044b\u0445 \u043a\u0440\u0443\u0433\u0430\u0445 \u0438 \u043d\u0430 \u0442\u0435\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0445 \u043a\u043e\u043d\u0444\u0435\u0440\u0435\u043d\u0446\u0438\u044f\u0445, \u043f\u043e\u0441\u0432\u044f\u0449\u0435\u043d\u043d\u044b\u0445 AI, \u043f\u043e\u044f\u0432\u0438\u043b\u043e\u0441\u044c \u043f\u043e\u043d\u044f\u0442\u0438\u0435 MLOps, \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u0431\u044b\u0441\u0442\u0440\u043e \u0437\u0430\u043a\u0440\u0435\u043f\u0438\u043b\u043e\u0441\u044c \u0432 \u043e\u0442\u0440\u0430\u0441\u043b\u0438 \u0438 \u0441\u0435\u0439\u0447\u0430\u0441 \u0440\u0430\u0437\u0432\u0438\u0432\u0430\u0435\u0442\u0441\u044f \u043a\u0430\u043a \u0441\u0430\u043c\u043e\u0441\u0442\u043e\u044f\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043d\u0430\u043f\u0440\u0430\u0432\u043b\u0435\u043d\u0438\u0435.\" \/>\n\t<meta name=\"robots\" content=\"max-image-preview:large\" \/>\n\t<meta name=\"author\" content=\"Yuri Gagarin\"\/>\n\t<link rel=\"canonical\" href=\"https:\/\/prohoster.info\/sk\/blog\/administrirovanie\/mlops-devops-v-mire-machine-learning\" \/>\n\t\t<meta name=\"generator\" content=\"All in One SEO (AIOSEO) 5.0.5\" \/>\n\t\t<meta property=\"og:locale\" content=\"sk_SK\" \/>\n\t\t<meta property=\"og:site_name\" content=\"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b\" \/>\n\t\t<meta property=\"og:type\" content=\"article\" \/>\n\t\t<meta property=\"og:title\" content=\"\ud83e\udd47MLOps: DevOps \u0432 \u043c\u0438\u0440\u0435 Machine Learning | ProHoster\" \/>\n\t\t<meta property=\"og:description\" content=\"\u0412 2018 \u0433\u043e\u0434\u0443 \u0432 \u043f\u0440\u043e\u0444\u0435\u0441\u0441\u0438\u043e\u043d\u0430\u043b\u044c\u043d\u044b\u0445 \u043a\u0440\u0443\u0433\u0430\u0445 \u0438 \u043d\u0430 \u0442\u0435\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0445 \u043a\u043e\u043d\u0444\u0435\u0440\u0435\u043d\u0446\u0438\u044f\u0445, \u043f\u043e\u0441\u0432\u044f\u0449\u0435\u043d\u043d\u044b\u0445 AI, \u043f\u043e\u044f\u0432\u0438\u043b\u043e\u0441\u044c \u043f\u043e\u043d\u044f\u0442\u0438\u0435 MLOps, \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u0431\u044b\u0441\u0442\u0440\u043e \u0437\u0430\u043a\u0440\u0435\u043f\u0438\u043b\u043e\u0441\u044c \u0432 \u043e\u0442\u0440\u0430\u0441\u043b\u0438 \u0438 \u0441\u0435\u0439\u0447\u0430\u0441 \u0440\u0430\u0437\u0432\u0438\u0432\u0430\u0435\u0442\u0441\u044f \u043a\u0430\u043a \u0441\u0430\u043c\u043e\u0441\u0442\u043e\u044f\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043d\u0430\u043f\u0440\u0430\u0432\u043b\u0435\u043d\u0438\u0435.\" \/>\n\t\t<meta property=\"og:url\" content=\"https:\/\/prohoster.info\/sk\/blog\/administrirovanie\/mlops-devops-v-mire-machine-learning\" \/>\n\t\t<meta property=\"og:image\" content=\"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg\" \/>\n\t\t<meta property=\"og:image:secure_url\" content=\"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg\" \/>\n\t\t<meta property=\"og:image:width\" content=\"350\" \/>\n\t\t<meta property=\"og:image:height\" content=\"350\" \/>\n\t\t<meta property=\"article:published_time\" content=\"2020-06-26T05:42:14+00:00\" \/>\n\t\t<meta property=\"article:modified_time\" content=\"2020-06-26T05:42:14+00:00\" \/>\n\t\t<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/prohoster\" \/>\n\t\t<meta property=\"article:author\" content=\"https:\/\/www.facebook.com\/prohoster\" \/>\n\t\t<!-- All in One SEO -->\n\n","aioseo_head_json":{"title":"\ud83e\udd47MLOps: DevOps vo svete Machine Learning | ProHoster","description":"V roku 2018 sa v profesion\u00e1lnych kruhoch a na tematick\u00fdch konferenci\u00e1ch venovan\u00fdch AI objavil pojem MLOps, ktor\u00fd sa r\u00fdchlo etabloval v odvetv\u00ed a v s\u00fa\u010dasnosti sa vyv\u00edja ako samostatn\u00e1 oblas\u0165.","canonical_url":"https:\/\/prohoster.info\/sk\/blog\/administrirovanie\/mlops-devops-v-mire-machine-learning","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":null,"og:locale":"sk_SK","og:site_name":"ProHoster | \u041a\u0443\u043f\u0438\u0442\u044c \u043d\u0430\u0434\u0435\u0436\u043d\u044b\u0439 \u0445\u043e\u0441\u0442\u0438\u043d\u0433 \u0434\u043b\u044f \u0441\u0430\u0439\u0442\u043e\u0432 \u0441 \u0437\u0430\u0449\u0438\u0442\u043e\u0439 \u043e\u0442 DDoS, VPS VDS \u0441\u0435\u0440\u0432\u0435\u0440\u044b","og:type":"article","og:title":"\ud83e\udd47MLOps: DevOps \u0432 \u043c\u0438\u0440\u0435 Machine Learning | ProHoster","og:description":"\u0412 2018 \u0433\u043e\u0434\u0443 \u0432 \u043f\u0440\u043e\u0444\u0435\u0441\u0441\u0438\u043e\u043d\u0430\u043b\u044c\u043d\u044b\u0445 \u043a\u0440\u0443\u0433\u0430\u0445 \u0438 \u043d\u0430 \u0442\u0435\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0445 \u043a\u043e\u043d\u0444\u0435\u0440\u0435\u043d\u0446\u0438\u044f\u0445, \u043f\u043e\u0441\u0432\u044f\u0449\u0435\u043d\u043d\u044b\u0445 AI, \u043f\u043e\u044f\u0432\u0438\u043b\u043e\u0441\u044c \u043f\u043e\u043d\u044f\u0442\u0438\u0435 MLOps, \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u0431\u044b\u0441\u0442\u0440\u043e \u0437\u0430\u043a\u0440\u0435\u043f\u0438\u043b\u043e\u0441\u044c \u0432 \u043e\u0442\u0440\u0430\u0441\u043b\u0438 \u0438 \u0441\u0435\u0439\u0447\u0430\u0441 \u0440\u0430\u0437\u0432\u0438\u0432\u0430\u0435\u0442\u0441\u044f \u043a\u0430\u043a \u0441\u0430\u043c\u043e\u0441\u0442\u043e\u044f\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043d\u0430\u043f\u0440\u0430\u0432\u043b\u0435\u043d\u0438\u0435.","og:url":"https:\/\/prohoster.info\/sk\/blog\/administrirovanie\/mlops-devops-v-mire-machine-learning","og:image":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:secure_url":"https:\/\/prohoster.info\/wp-content\/uploads\/2021\/11\/logo-350.jpg","og:image:width":350,"og:image:height":350,"article:published_time":"2020-06-26T05:42:14+00:00","article:modified_time":"2020-06-26T05:42:14+00:00","article:publisher":"https:\/\/www.facebook.com\/prohoster","article:author":"https:\/\/www.facebook.com\/prohoster"},"aioseo_meta_data":{"post_id":"86523","title":null,"description":null,"keywords":null,"keyphrases":null,"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":null,"og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"","isEnabled":true},"graphs":[]},"schema_type":null,"schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":null,"robots_max_videopreview":null,"robots_max_imagepreview":"large","priority":null,"frequency":null,"local_seo":null,"seo_analyzer_scan_date":null,"breadcrumb_settings":null,"limit_modified_date":false,"reviewed_by":null,"ai":null,"created":"2021-02-28 14:13:23","updated":"2022-09-28 09:22:54","focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/posts\/86523","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/comments?post=86523"}],"version-history":[{"count":0,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/posts\/86523\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/media\/86524"}],"wp:attachment":[{"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/media?parent=86523"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/categories?post=86523"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/prohoster.info\/sk\/wp-json\/wp\/v2\/tags?post=86523"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}