I-InterSystems IRIS - iqonga lexesha langempela le-AI / ML

Umbhali: USergey Lukyanchikov, injineli yomcebisi kwi-InterSystems

Ixesha langempela le-AI/ML kwikhompyutha yeefowuni

Masiqale ngemizekelo evela kumava eNzululwazi yeDatha kwi-InterSystems:

• I-portal yomthengi elayishiweyo iqhagamshelwe kwinkqubo yengcebiso ye-intanethi. Kuya kubakho ukuhlengahlengiswa kweentengiso kuwo wonke umnatha wokuthengisa (umzekelo, endaweni yomgca "we-flat" wokunyusa, i-matrix "yecandelo-amaqhinga" iya kusetyenziswa ngoku). Kwenzeka ntoni kwiinjini zokucebisa? Kwenzeka ntoni ekungenisweni nasekuhlaziyweni kwedatha kwi-injini yokucebisa (umthamo wedatha yokufaka unyuke ngamaxesha angama-25000)? Kwenzeka ntoni ekuphuhlisweni kweengcebiso (isidingo sokunciphisa umda wokucoca wemigaqo yokuncoma ngokuphindwe ngewaka ngenxa yokunyuka kwewaka kwinani labo kunye "noluhlu")?
• Kukho inkqubo yokubeka esweni ukuba nokwenzeka kweziphene eziphuhlayo kumacandelo ezixhobo. Inkqubo yokulawula inkqubo ezenzekelayo idityaniswe nenkqubo yokubeka iliso, ihambisa amawaka eeparamitha zenkqubo yeteknoloji nganye yesibini. Kwenzeka ntoni kwisistim yokubeka iliso ebisebenza ngaphambili “kwiisampulu zezandla” (ingaba iyakwazi ukubonelela ngohlolo lokunokwenzeka lwesibini ngesibini)? Kuya kwenzeka ntoni ukuba ibhloko entsha yeekholamu ezingamakhulu aliqela ibonakala kwidatha yegalelo kunye nokufundwa kwabenzi boluvo bongezwe mva nje kwinkqubo yolawulo lwenkqubo (ingaba kuyafuneka kwaye ixesha elingakanani ukumisa inkqubo yokubeka iliso ukubandakanya idatha evela kuluvo olutsha kuhlalutyo. )?
• Isethi yeendlela ze-AI / ML (izincomo, ukubeka iliso, ukubikezela) zenziwe ezisebenzisa iziphumo zomsebenzi womnye nomnye. Zingaphi iiyure zomntu ezifunekayo nyanga nganye ukulungelelanisa ukusebenza kolu bunzima kwiinguqu kwidatha yegalelo? Yintoni "ukucotha" ngokubanzi xa kuxhaswe yinkqubo yokwenza izigqibo zolawulo (ukuphindaphinda kokwenzeka kolwazi olutsha oluxhasayo kuyo ngokunxulumene nokuphindaphinda ukwenzeka kwedatha entsha yegalelo)?

Ukushwankathela le mizekelo kunye neminye emininzi, sifikile ekuqulunqweni kwemingeni evelayo xa sifudukela ekusebenziseni umatshini wokufunda kunye neendlela zobukrelekrele bokwenziwa ngexesha langempela:

• Ngaba siyaneliseka ngesantya sokudala kunye nokulungelelaniswa (kwimeko eguqukayo) yophuhliso lwe-AI/ML kwinkampani yethu?
• Zingakanani izisombululo ze-AI/ML esizisebenzisayo zixhasa ulawulo lweshishini ngexesha lokwenyani?
• Ngaba izisombululo ze-AI / ML esizisebenzisayo ziyakwazi ukuzimela (ngaphandle kwabaphuhlisi) ukulungelelanisa utshintsho kwiidatha kunye neendlela zokulawula ishishini?

Inqaku lethu liyinkcazo epheleleyo yezakhono ze-InterSystems iqonga le-IRIS ngokubhekiselele kwinkxaso yehlabathi jikelele yokuthunyelwa kweendlela ze-AI / ML, indibano (ukuhlanganiswa) yezisombululo ze-AI / ML, kunye noqeqesho (uvavanyo) lwe-AI / ML izisombululo kwi-intensive imisinga yedatha. Siza kujonga uphando lweemarike, izifundo ze-AI / ML zezisombululo, kunye neengcamango zento esiyibiza ngokuba yi-real-time AI / ML iqonga kweli nqaku.

Into esiyaziyo kuphando: usetyenziso lwexesha lokwenyani

Iziphumo uphandoiqhutywe phakathi kweengcali ze-IT eziphantse zibe ngama-800 ngo-2019 nguLightbend, zizithethele:

Umzobo 1 Abathengi abakhokelayo bedatha yexesha langempela

Makhe sicaphule amaqhekeza abalulekileyo engxelo ngeziphumo zolu phando kuguqulo lwethu:

"... Iindlela zokuthandwa kwezixhobo zokudibanisa imilambo yedatha kwaye, ngelo xesha, ukuxhasa i-computing kwiikhonteyina zibonelela ngempendulo ye-synergistic kwisicelo semarike yesiphakamiso esiphendulayo, esinengqiqo, esinamandla sezisombululo ezisebenzayo. Idatha yokusasaza idlulisela ulwazi ngokukhawuleza kunedatha yepakethe yemveli. Ukongezwa kule nto kukukwazi ukusebenzisa ngokukhawuleza iindlela zokubala, ezifana, umzekelo, iingcebiso ezisekelwe kwi-AI / ML, ukudala inzuzo yokukhuphisana ngokunyuka kokwaneliseka kwabathengi. Umdyarho we-agility nawo uchaphazela zonke iindima kwi-paradigm ye-DevOps - ukwenza uphuhliso lwesicelo kunye nokusasazwa ngokufanelekileyo. … Amakhulu asibhozo anesine iingcali ze-IT zinike ulwazi malunga nokusetyenziswa kokuhamba kwedatha kwimibutho yazo. Abaphenduli ubukhulu becala babekwe kumazwe aseNtshona (i-41% eYurophu kunye ne-37% kuMntla Melika) kwaye baphantse basasazwa ngokulinganayo phakathi kweenkampani ezincinci, eziphakathi kunye nezinkulu. ...

... Ubukrelekrele bokwenziwa abuyontsomi. Amashumi amahlanu anesibhozo ekhulwini abo sele besebenzisa ukuqhutyelwa kwedatha kwimveliso ye-AI / ML izicelo ziqinisekisa ukuba ukusebenzisa kwabo i-AI / ML kuya kubona ukukhula okukhulu kunyaka ozayo (xa kuthelekiswa nezinye izicelo).

• Ngokwabaninzi abaphenduliweyo, ukusetyenziswa kwedatha yedatha kwiimeko ze-AI / ML kuya kubona ukukhula okukhulu kunyaka ozayo.
• Izicelo kwi-AI/ML aziyi kukhula kuphela ngenxa yeentlobo ezintsha zeemeko, kodwa nangenxa yeemeko zemveli apho idatha yexesha langempela isetyenziswa kakhulu.
• Ukongeza kwi-AI / ML, inqanaba lentshiseko phakathi kwabasebenzisi bemibhobho yedatha ye-IoT iyamangalisa - i-48% yalabo sele bedibanise idatha ye-IoT bathi ukuphunyezwa kwemeko kule datha kuya kubona ukwanda okukhulu kwixesha elizayo. ..."

Kolu phando lunomdla kakhulu, kucacile ukuba umbono wokufunda koomatshini kunye neemeko zobukrelekrele bokwenziwa njengeenkokeli ekusetyenzisweni kwemijelo yedatha sele "isendleleni." Kodwa uqwalaselo olubaluleke ngokulinganayo kumbono wexesha lokwenyani le-AI/ML ngelensi ye-DevOps: apha sinokuqala ukuthetha ngoguqulo lwenkcubeko esele igqwesile “ye-AI/ML elahlayo eneseti yedatha efikeleleka ngokupheleleyo.”

Ixesha langempela le-AI / ML ingcamango yeqonga

Indawo enye eqhelekileyo yesicelo se-AI/ML yexesha langempela lulawulo lwenkqubo kwimveliso. Ukusebenzisa umzekelo wakhe kunye nokuqwalasela iingcamango zangaphambili, siya kuqulunqa ingcamango yeqonga le-AI / ML yexesha langempela.
Ukusetyenziswa kobukrelekrele bokwenziwa kunye nokufunda koomatshini kulawulo lwenkqubo kuneempawu ezininzi:

• Idatha kwimeko yenkqubo yetekhnoloji yamkelwa ngamandla: ngokuphindaphinda okuphezulu kunye noluhlu olubanzi lweeparamitha (ukuya kuthi ga kumawaka amaxabiso eparameter ahanjiswa ngomzuzwana ukusuka kwinkqubo yolawulo lwenkqubo)
• Idatha malunga nokuchongwa kweziphene, kungakhankanywa idatha ekuphuhlisweni kwayo, ngokuchaseneyo, inqabile kwaye ingaqhelekanga, ibonakaliswe ngokulinganisa okwaneleyo kweziphene kunye nokubekwa kwendawo ngexesha (elihlala limelwe ngamarekhodi ephepha)
• Ukusuka kwimbono ebonakalayo, kuphela "ifestile yokufaneleka" yomthombo wedatha ekhoyo yoqeqesho kunye nokusebenzisa iimodeli, ebonisa amandla enkqubo yetekhnoloji ngexesha elifanelekileyo lokutyibilika eliphela ngexabiso lokugqibela lokufunda iparameters zenkqubo.

Ezi mpawu ziyasinyanzela, ukongeza ekufumaneni nasekuqhubeni okusisiseko ngexesha lokwenyani "lophawu lwegalelo elibanzi" olusuka kwinkqubo yetekhnoloji, ukwenza (ngaxeshanye) isicelo, uqeqesho kunye nolawulo lomgangatho weziphumo ze-AI / Iimodeli zeML - nazo ngexesha langempela. "Isakhelo" esiyimodeli yethu "esibonayo" kwifestile ye-sliding ye-relevance iguquguquka rhoqo - kwaye kunye nayo, umgangatho weziphumo zomsebenzi weemodeli ze-AI / ML eziqeqeshwe kwenye ye "isakhelo" kwixesha elidlulileyo litshintsha kwakhona. . Ukuba umgangatho weziphumo zomsebenzi weemodeli ze-AI/ML uyawohloka (umzekelo: ixabiso lempazamo yohlelo lwe "alarm-norm" liye ngaphaya kwemida esiyichazileyo), uqeqesho olongezelelweyo lweemodeli kufuneka luqaliswe ngokuzenzekelayo. "isakhelo" sangoku ngakumbi - kunye nokukhethwa kwexesha lokuqalisa uqeqesho olongezelelweyo lweemodeli kufuneka kuthathelwe ingqalelo indlela ixesha loqeqesho ngokwalo, kunye nokutshintsha komgangatho womsebenzi wenguqu yangoku yeemodeli (ukusukela iinguqulelo zangoku zeemodeli ziyaqhubeka zisetyenziswa ngelixa iimodeli ziqeqeshwa, kwaye de iinguqulelo zabo "ezisanda kuqeqeshwa" zenziwe).

I-InterSystems IRIS inobuchule obuphambili beqonga ukwenza i-AI/ML izisombululo zolawulo lwenkqubo yexesha langempela. Ezi zakhono zinokwahlulwa zibe ngamaqela amathathu aphambili:

• Ukuthunyelwa ngokuqhubekayo (Ukuthunyelwa ngokuqhubekayo / ukuhanjiswa, iCD) yeendlela ezintsha okanye ezilungelelanisiweyo ze-AI / ML kwisisombululo esivelisayo esisebenza ngexesha langempela kwi-InterSystems IRIS iqonga.
• Ukuhlanganiswa okuqhubekayo (CI) kwisisombululo esisodwa esivelisayo senkqubo yeteknoloji engenayo imijelo yedatha yedatha, imigca yedatha yesicelo / uqeqesho / ulawulo lomgangatho weendlela ze-AI / ML kunye notshintshiselwano lwedatha / ikhowudi / izenzo zokulawula kunye neendawo zokumisela i-mathematika, ezicwangciswe ngexesha langempela. iqonga InterSystems IRIS
• Uqeqesho oluqhubekayo (lokuzimela) (uQeqesho oluqhubekayo, i-CT) yeendlela ze-AI / ML, ezenziwa kwiindawo zokulinganisa izibalo usebenzisa idatha, ikhowudi kunye nezenzo zokulawula ("izigqibo ezenziwe") ezithunyelwa yi-InterSystems IRIS iqonga.

Ukuhlelwa kwezakhono zeqonga ngokumalunga nokufunda koomatshini kunye nobukrelekrele bokwenziwa ngokuchanekileyo kula maqela akuyongozi. Makhe sicaphule i-methodological upapasho UGoogle, obonelela ngesiseko sengqiqo kolu lwahlulo, kwinguqulelo yethu:

“... Ingqikelelo ye-DevOps, edumileyo kule mihla, ibandakanya uphuhliso kunye nokusebenza kweenkqubo zolwazi olukhulu. Iinzuzo zokuphumeza le ngcamango kukunciphisa ixesha lemijikelezo yophuhliso, ukuhanjiswa ngokukhawuleza kophuhliso, kunye nokuguquguquka ekucwangcisweni kokukhululwa. Ukufezekisa ezi zibonelelo, i-DevOps ibandakanya ukuphunyezwa kweendlela ezimbini:

• Udibaniso oluqhubekayo (CI)
• Unikezelo oluqhubekayo (CD)

Ezi zenzo zikwasebenza kwiiplatifti ze-AI/ML zokuqinisekisa indibano ethembekileyo nesebenzayo yezisombululo ze-AI/ML ezinemveliso.

Amaqonga e-AI/ML ahluka kwezinye iinkqubo zolwazi kule miba ilandelayo:

• Ubuchule beqela: Xa udala isisombululo se-AI / ML, iqela lidla ngokubandakanya izazinzulu zedatha okanye iingcali "zezemfundo" kwinkalo yophando lwedatha eqhuba uhlalutyo lwedatha, ukuphuhlisa nokuvavanya imizekelo. La malungu eqela asenokungabi ngabaphuhlisi bekhowudi abanobuchule.
• Uphuhliso: Iinjini ze-AI/ML ziyalingwa ngokwendalo. Ukuze ucombulule ingxaki ngeyona ndlela isebenzayo, kuyimfuneko ukuhamba ngeendlela ezahlukeneyo zokudibanisa igalelo, i-algorithms, iindlela zokwenza imodeli kunye neeparitha zemodeli. Ubunzima bophando olunjalo luxhomekeke ekulandeleni "into esebenzayo / engazange isebenze", ukuqinisekisa ukuphindaphinda kweziqephu, ukuveliswa ngokubanzi kophuhliso lokuphunyezwa ngokuphindaphindiweyo.
• Uvavanyo: Ukuvavanya iinjini ze-AI/ML kufuna uluhlu olubanzi lovavanyo kunolunye uphuhliso. Ukongeza kwiyunithi esemgangathweni kunye neemvavanyo zokudibanisa, ukuqinisekiswa kwedatha kunye nomgangatho weziphumo zokusebenzisa imodeli ekuqeqesheni nasekulawuleni iisampuli zivavanywa.
• Ukusasazwa: Ukuhanjiswa kwezisombululo ze-AI / ML akukhawulelwanga kwiinkonzo zokuqikelela ezisebenzisa imodeli eqeqeshwe kanye. Izisombululo ze-AI/ML zakhiwe malunga nemibhobho enamanqanaba amaninzi eyenza uqeqesho oluzenzekelayo lwemodeli kunye nokusetyenziswa. Ukusasazwa kwemibhobho enjalo kubandakanya ukuzenzekela izenzo ezingeyomfuneko ngokwesiko ezenziwa ngesandla zizazinzulu zedatha ukuze zikwazi ukuqeqesha nokuvavanya imifuziselo.
• Imveliso: Iinjini ze-AI / ML zingakwazi ukuvelisa imveliso kungekhona nje ngenxa yeprogram engasebenziyo, kodwa nangenxa yokutshintsha rhoqo kwedatha yegalelo. Ngamanye amazwi, ukusebenza kweendlela ze-AI/ML zinokuthotywa ngenxa yoluhlu olubanzi lwezizathu kunokwenziwa kophuhliso oluqhelekileyo. Okukhokelela kwisidingo sokubeka iliso (kwi-intanethi) ukusebenza kweenjini zethu ze-AI / ML, kunye nokuthumela izilumkiso okanye ukugatya iziphumo ukuba izikhombisi zokusebenza azihlangabezani nokulindela.

Iiplatifti ze-AI / ML zifana nezinye iinkqubo zolwazi apho zombini zifuna ukuhlanganiswa kwekhowudi eqhubekayo kunye nokulawulwa kwenguqulo, uvavanyo lweeyunithi, uvavanyo lokudibanisa, kunye nophuhliso oluqhubekayo. Nangona kunjalo, kwimeko ye-AI/ML, kukho umahluko obalulekileyo:

• I-CI (i-Continuous Integration) ayisekho kuphela ekuvavanyeni nasekuqinisekiseni ikhowudi yamacandelo asetyenzisiweyo - iquka ukuvavanya nokuqinisekisa idatha kunye neemodeli ze-AI / ML.
• I-CD (Ukuhanjiswa ngokuqhubekayo / ukuhanjiswa, ukuthunyelwa okuqhubekayo) akuphelelanga ekubhaleni nasekukhupheni iiphakheji okanye iinkonzo, kodwa kuthetha iqonga lokuqulunqa, ukuqeqeshwa kunye nokusetyenziswa kwezixazululo ze-AI / ML.
• I-CT (uQeqesho oluqhubekayo, uqeqesho oluqhubekayo) yinto entsha [approx. umbhali wenqaku: into entsha ngokumalunga nombono wendabuko we-DevOps, apho i-CT, njengomthetho, uVavanyo oluqhubekayo], oluhambelana neeplatifti ze-AI / ML, ezijongene nolawulo oluzimeleyo lweendlela zokuqeqesha kunye nokusebenzisa i-AI. / iimodeli zeML. ..."

Singatsho ukuba ukufundwa komatshini kunye nobukrelekrele bokwenziwa obusebenza kwidatha yexesha langempela kufuna iseti ebanzi yezixhobo kunye nobuchule (ukusuka kuphuhliso lwekhowudi ukuya kwi-orchestration yemeko yemodel yemathematika), ukudityaniswa ngokusondeleyo phakathi kwazo zonke iindawo ezisebenzayo kunye nezifundo, intlangano esebenzayo ngakumbi yabantu kunye izixhobo zoomatshini.

Imeko yexesha langempela: ukuqaphela ukuphuhliswa kweziphene kwiimpompo zokutya

Ukuqhubeka nokusebenzisa indawo yolawulo lwenkqubo njengomzekelo, qwalasela ingxaki ethile (esele siyikhankanyile ekuqaleni): kufuneka sinikeze iliso lexesha langempela lophuhliso lweziphene kwiimpompo ezisekelwe ekuhambeni kwemilinganiselo yepharamitha yenkqubo. kunye neengxelo ezivela kubasebenzi bokulungisa malunga neziphene ezichongiweyo.

Umzobo 2 Ukuqulunqwa kwengxaki yokubeka iliso kuphuhliso lweziphene

Uphawu lwemisebenzi emininzi eyenziwa ngolu hlobo xa kusenziwa lo msebenzi kukuba ukwamkelwa kwedatha rhoqo kunye nobuchule bokufumana iirisithi (i-APCS) kufuneka kuqwalaselwe ngokweepisodic nezehlo ezingaqhelekanga (kunye nokubhaliswa) kweziphene zeentlobo ngeentlobo. Ngamanye amazwi: idatha evela kwinkqubo yokulawula inkqubo ifika kanye okwesibini, ichanekile kwaye ichanekile, kwaye amanqaku malunga neziphene zenziwe ngepensile yeekhemikhali ebonisa umhla kwincwadana eqhelekileyo kwindawo yokusebenzela (umzekelo: "12.01 - ukuvuza kwikhava ukusuka kwicala le-3rd bear").

Ngaloo ndlela, sinokuncedisa ukuqulunqwa kwengxaki ngokunciphisa okubalulekileyo okulandelayo: sine "ileyibhile" enye kuphela yesiphene sohlobo oluthile (oko kukuthi, umzekelo wesiphene sohlobo oluthile lumelwe yidatha evela kwinkqubo yokulawula inkqubo inkqubo ngomhla othile - kwaye asinayo eminye imizekelo yesiphene solu hlobo). Lo mda ngokukhawuleza usithatha ngaphaya kwendawo yokufunda koomatshini beklasi (ukufunda okugadiweyo), apho kufuneka kubekho "iithegi" ezininzi.

Umzobo 3 Ukucaciswa komsebenzi wokubeka iliso kuphuhliso lweziphene

Ngaba ngandlel’ ithile singakwazi “uphindaphinda” “ithegi” ekukuphela kwayo esinayo? Ewe singakwazi. Imeko yangoku yempompo ibonakala ngezinga lokufana neziphene ezibhalisiweyo. Ngaphandle kokusetyenziswa kweendlela zobungakanani, kwinqanaba lokujonga okubonakalayo, ngokujonga amandla amaxabiso edatha evela kwinkqubo yolawulo lwenkqubo, unokufunda okuninzi:

Umzobo we-4 Ukutshintsha imeko yempompo ngokuchasene nemvelaphi "yophawu" lwesiphene sohlobo olunikiweyo.

Kodwa imbono yokubona (ubuncinci okwangoku) ayiyona i-generator efanelekileyo kakhulu "yethegi" kwimeko yethu eguqukayo ngokukhawuleza. Siza kuvavanya ukufana kwemeko yempompo yangoku kwiziphene ezixeliweyo kusetyenziswa uvavanyo lwamanani.

Umzobo 5 Ukufaka uvavanyo lwamanani kwidatha engenayo ngokuchasene nemvelaphi yesiphene "ileyibhile"

Uvavanyo lwamanani lumisela ukuba kunokwenzeka ukuba iirekhodi ezinamaxabiso eeparamitha zenkqubo yetekhnoloji kwi "flow-packet" efunyenwe kwinkqubo yolawulo lwenkqubo iyafana neerekhodi ze "tag" yesiphene sohlobo oluthile. Ixabiso elinokwenzeka (isalathisi sokufana kwezibalo) esibalwe ngenxa yokusetyenziswa kovavanyo lwamanani luguqulelwa kwixabiso le-0 okanye i-1, ibe "ileyibhile" yokufunda ngomatshini kwirekhodi nganye ethile kwipakethi ehlolwe ukufana. Oko kukuthi, emva kokucubungula iphakheji entsha efunyenweyo yeerekhodi zombuso wempompo kunye novavanyo lwamanani, sinethuba lokuba (a) songeze le phakheji kwiseti yoqeqesho loqeqesho lwemodeli ye-AI / ML kunye (b) nokulawula umgangatho we uguqulelo lwangoku lomfuziselo xa uyisebenzisa kule mpahla.

Umzobo 6 Ukusebenzisa imodeli yokufunda yomatshini kwidatha engenayo ngokuchasene nemvelaphi yesiphene "ileyibhile"

Kwenye yangaphambili yethu iiwebinars Sibonisa kwaye sicacisa indlela iqonga le-InterSystems IRIS likuvumela ukuba uphumeze nayiphi na indlela ye-AI/ML ngendlela yokuqhuba ngokuqhubekayo iinkqubo zoshishino ezibeka iliso ekuthembekeni kweziphumo zemodeli kunye nokulungelelanisa iiparitha zemodeli. Xa siphumeza i-prototype yemeko yethu kunye neepompo, sisebenzisa yonke i-InterSystems IRIS yokusebenza evezwe ngexesha le-webinar - ukuphunyezwa kwinkqubo yokuhlaziya njengenxalenye yesisombululo sethu kungekhona ukufunda okuphantsi kweklasi, kodwa kunokomeleza ukufunda, okulawula ngokuzenzekelayo ukhetho lweemodeli zoqeqesho. . Isampulu yoqeqesho iqulethe iirekhodi apho "imvumelwano yokubhaqwa" ivela emva kokusetyenziswa kokubini uvavanyo lwamanani kunye noguqulelo lwangoku lwemodeli - oko kukuthi, zombini uvavanyo lwamanani (emva kokuguqula isalathiso sokufana kwi-0 okanye i-1) kwaye imodeli yavelisa umphumo. kwezo rekhodi 1. Ngexesha loqeqesho olutsha lwemodeli, ngexesha lokuqinisekiswa kwayo (imodeli esanda kuqeqeshwa isetyenziswa kwisampulu yoqeqesho lwayo, kunye nokusetyenziswa kwangaphambili kovavanyo lweenkcukacha-manani kuyo), iirekhodi "ezingazange zigcine" iziphumo ze-1 emva kokucubungula. ngovavanyo lweenkcukacha-manani (ngenxa yokubakho rhoqo kuqeqesho isampuli yeerekhodi ezivela “kwileyibhile” yokuqala yesiphene), ziyasuswa kwiseti yoqeqesho, kwaye inguqulelo entsha yemodeli ifunda “kwileyibhile” ye isiphene kunye neerekhodi "ezigciniweyo" ezivela kumsinga.

Umzobo we-7 Robotization ye-AI / ML computing kwi-InterSystems IRIS

Ukuba kukho imfuno yohlobo "loluvo lwesibini" kumgangatho wokufunyanwa okufunyenwe ngexesha lokubala kwendawo kwi-InterSystems IRIS, inkqubo yomcebisi idalwa ukwenza uqeqesho kunye nokusetyenziswa kweemodeli kwi-dataset yokulawula usebenzisa iinkonzo zefu (umzekelo, iMicrosoft IAzure, iiNkonzo zeWebhu zeAmazon, iPlatifomu yeLifu likaGoogle, njl.):

Umzobo 8 Uluvo lweSibini oluvela kuMicrosoft Azure oluququzelelwe yi-InterSystems IRIS

Iprototype yemeko yethu kwi-InterSystems IRIS yenzelwe njengenkqubo esekelwe kwi-arhente yeenkqubo zokuhlalutya ezisebenzisana nesixhobo sesixhobo (ipompo), imodeli yeemathematika (Python, R kunye noJulia), kunye nokuqinisekisa ukuzifundela kwabo bonke ababandakanyekayo kwi-AI / Iindlela ze-ML - kwimijelo yedatha yexesha langempela.

Umzobo 9 Umsebenzi oyintloko wexesha langempela le-AI / ML isisombululo kwi-InterSystems IRIS

Iziphumo ezisebenzayo zeprototype yethu:

• Isiphene sesampulu esiqatshelwe yimodeli (ngoJanuwari 12):

• Isiphene esiphuhlayo esiqatshelweyo yimodeli engazange ifakwe kwisampuli (ngoSeptemba 11, isiphene ngokwaso ichongiwe liqela lokulungisa kuphela iintsuku ezimbini kamva, ngoSeptemba 13):

Ukulinganisa kwiidatha zangempela eziqulethe iziqendu ezininzi zesiphene esifanayo zibonise ukuba isisombululo sethu, esiphunyezwe kwi-InterSystems IRIS iqonga, lisivumela ukuba sibone ukuphuhliswa kweziphene zolu hlobo iintsuku ezininzi ngaphambi kokuba zibonwe liqela lokulungisa.

I-InterSystems IRIS - iqonga lekhompiyutha le-AI/ML yexesha langempela jikelele

Iqonga le-InterSystems IRIS lenza lula uphuhliso, ukuthunyelwa kunye nokusebenza kwezisombululo zedatha yexesha langempela. I-InterSystems IRIS iyakwazi ukwenza ngaxeshanye ukusetyenzwa kwedatha kunye nokuhlalutya idatha; xhasa iimboniselo zedatha ezilungelelanisiweyo ngokwemifuziselo emininzi (kubandakanywa ubudlelwane, uluhlu, into kunye noxwebhu); sebenza njengeqonga lokudibanisa uluhlu olubanzi lwemithombo yedatha kunye nezicelo zomntu ngamnye; nikezela ngohlalutyo lwexesha lokwenyani eliphezulu kwidatha eyakhiweyo nengalungiswanga. I-InterSystems IRIS ikwabonelela ngeendlela zokusebenzisa izixhobo zokuhlalutya zangaphandle kwaye ivumela indibaniselwano eguquguqukayo yokubamba kwilifu nakwiseva yendawo.

Izicelo ezakhiwe kwi-InterSystems iqonga le-IRIS zisasazwa kuwo wonke amashishini ahlukeneyo, ukunceda iinkampani ukuba zibone inzuzo ebalulekileyo yezoqoqosho kwimbono yesicwangciso kunye nokusebenza, ukwandisa ukuthathwa kwezigqibo ezinolwazi kunye nokuvala izithuba phakathi kwesiganeko, uhlalutyo kunye nesenzo.

Umzobo we-10 we-InterSystems i-IRIS uyilo kumxholo wexesha langempela le-AI / ML

Njengomzobo wangaphambili, umzobo ongezantsi udibanisa "inkqubo yokulungelelanisa" entsha (i-CD / CI / CT) kunye nomzobo wokuhamba kolwazi phakathi kwezinto zokusebenza zeqonga. Ukubonwa kuqala ngeCD ye-macromechanism kwaye kuqhubeke nge-macromechanisms CI kunye ne-CT.

Umzobo we-11 umzobo wolwazi uhamba phakathi kwezinto ze-AI / ML ze-InterSystems IRIS iqonga

Ingundoqo ye-CD ye-CD kwi-InterSystems IRIS: abasebenzisi beqonga (abaphuhlisi bezisombululo ze-AI / ML) bahlengahlengise ezikhoyo kunye / okanye benze uphuhliso olutsha lwe-AI / ML usebenzisa umhleli wekhowudi ekhethekileyo yeendlela ze-AI / ML: Jupyter (igama elipheleleyo: Jupyter Notebook; ngobufutshane, amaxwebhu enziwe kulo mhleli nawo ngamanye amaxesha abizwa). Kwi-Jupyter, umphuhlisi unethuba lokubhala, ukulungisa iimpazamo kunye nokuqinisekisa ukusebenza (kubandakanywa nokusetyenziswa kwemizobo) yophuhliso oluthile lwe-AI / ML ngaphambi kokuba ibekwe ("ifakwe") kwi-InterSystems IRIS. Kucacile ukuba uphuhliso olutsha oludalwe ngolu hlobo luya kufumana kuphela ukulungiswa kwesiseko (ekubeni, ngokukodwa, iJupyter ayisebenzi kunye nemijelo yedatha yexesha langempela) - oku kulandelelwano lwezinto, kuba umphumo oyintloko wophuhliso kwiJupyter. sisiqinisekiso sokusebenza okusisiseko kwendlela eyahlukileyo ye-AI/ML (“ibonisa iziphumo ezilindelekileyo kwisampulu yedatha”). Ngokufanayo, indlela esele ibekwe eqongeni (jonga le mikhro-mechanism ilandelayo) phambi kokuba kulungiswe ingxaki kwiJupyter inokufuna “ukubuyisela umva” kwifom “yangaphambili” (ukufunda idatha kwiifayile, ukusebenza ngedatha nge-xDBC endaweni yeetafile, ukusebenzisana ngokuthe ngqo kunye nehlabathi jikelele - uluhlu lwedatha ye-multidimensional InterSystems IRIS - njl.).

Inkalo ebalulekileyo yokuphunyezwa kweCD kwi-InterSystems IRIS: ukuhlanganiswa kwe-bidirectional kuye kwaphunyezwa phakathi kweqonga kunye neJupyter, ukuvumela umxholo wePython, uR noJulia ukuba udluliselwe kwiqonga (kwaye, emva koko, ukucutshungulwa kwiqonga) (zonke ezintathu ziyiprogram iilwimi kwiilwimi ezinomthombo ovulekileyo ohambelanayo nomthombo wemodeli yezibalo). Ngaloo ndlela, abaphuhlisi bomxholo we-AI / ML banethuba lokuqhuba "ukuthunyelwa ngokuqhubekayo" kwalo mxholo eqongeni, besebenza kumhleli wabo weJupyter owaziwayo, kunye namathala eencwadi aqhelekileyo akhoyo kwiPython, R, Julia, kunye nokwenza ukulungiswa kwesiseko (ukuba kuyimfuneko) ngaphandle kweqonga .

Masiqhubele phambili kwindlela ye-CI macro kwi-InterSystems IRIS. Umzobo ubonisa inkqubo ye-macro "yexesha lokwenyani irobhothi" (ubunzima bedatha yedatha, iinkqubo zoshishino kunye neengceba zekhowudi ezicwangciswe ngabo kwiilwimi zemathematika kunye ne-ObjectScript - ulwimi lwemveli lophuhliso lwe-InterSystems IRIS). Umsebenzi wale nkqubo ye-macro kukugcina imigca yedatha eyimfuneko ekusebenzeni kweendlela ze-AI / ML (esekelwe kwimithombo yedatha edluliselwe kwiqonga ngexesha langempela), yenza izigqibo malunga nokulandelelana kwesicelo kunye "ne-assortment" ye-AI / Iindlela ze-ML (zikwayi "algorithms yemathematika", "imodeli", njl. njl. - zinokubizwa ngokwahlukileyo ngokuxhomekeke kwiinkcukacha zokuphunyezwa kunye nokukhethwa kwesigama), gcina izakhiwo zedatha zihlaziyiwe ukuhlalutya iziphumo zomsebenzi we-AI / Iindlela ze-ML (iityhubhu, iitafile, uluhlu lwedatha ye-multidimensional, njl. njl. - kwiingxelo, iideshibhodi, njl.).

Inkalo ebalulekileyo yokuphunyezwa kwe-CI ngokukodwa kwi-InterSystems IRIS: ukudityaniswa kwe-bidirectional kuye kwaphunyezwa phakathi kweqonga kunye neendawo zokubunjwa kweemathematika, kukuvumela ukuba uqhube umxholo obanjwe kwiqonga ePython, R noJulia kwiindawo zabo kunye nokubuyisela ukuphunyezwa. iziphumo. Oku kudityaniswa kuphunyezwa zombini "kwimo yesiphelo" (oko kukuthi, umxholo we-AI/ML uqulunqwe njengekhowudi ye-ObjectScript eyenza iminxeba kwindalo esingqongileyo) kunye "kwimowudi yenkqubo yeshishini" (oko kukuthi, umxholo we-AI/ML uqulunqwe njengenkqubo yoshishino usebenzisa umhleli womzobo, okanye ngamanye amaxesha usebenzisa iJupyter, okanye usebenzisa i-IDE-IRIS Studio, Eclipse, Visual Studio Code). Ukufumaneka kweenkqubo zoshishino zokuhlela kwiJupyter kubonakaliswa ngoqhagamshelwano phakathi kwe-IRIS kwinqanaba le-CI kunye neJupyter kwinqanaba leCD. Isishwankathelo esinenkcukacha ezithe vetshe zokudityaniswa neemeko zemodeli yezibalo zinikiwe ngezantsi. Ngeli nqanaba, ngokombono wethu, kukho zonke izizathu zokuqinisekisa ukuba iqonga linazo zonke izixhobo eziyimfuneko zokuphumeza "ukuhlanganiswa okuqhubekayo" kophuhliso lwe-AI / ML (ephuma "ekuhanjisweni okuqhubekayo") kwiisombululo zangempela ze-AI / ML.

Kwaye eyona ndlela iphambili ye-macro: CT. Ngaphandle kwayo, akuyi kubakho iqonga le-AI / ML (nangona "ixesha langempela" liya kuphunyezwa ngeCD / CI). Ingundoqo ye-CT ngumsebenzi weqonga kunye "nobugcisa" bokufunda ngomatshini kunye nobukrelekrele bokufakelwa ngokuthe ngqo kwiiseshoni zokusebenza zeendawo zemodeli yeemathematika: iimodeli, iitafile zokusabalalisa, ii-matrix vectors, iileya ze-neural networks, njl. Lo "msebenzi", kwiimeko ezininzi, ubandakanya ukudala izinto zakudala ezikhankanyiweyo kwindawo (kwimeko yeemodeli, umzekelo, "indalo" ibandakanya ukusetwa kwemodeli kunye nokukhetha okulandelayo kwamaxabiso eeparamitha zayo - oko kubizwa ngokuba "luqeqesho" lwemodeli), isicelo sabo (kwimodeli: ukubala ngoncedo lwabo "imodeli" yexabiso lezinto ekujoliswe kuzo - uqikelelo, ubulungu becandelo, ukubakho kwesiganeko, njl. ii-artifacts ezidalwe kwaye zisetyenzisiwe (umzekelo, ukucacisa kwakhona isethi yeenguqu zegalelo zemodeli ngokusekelwe kwiziphumo zesicelo - ukwenzela ukuba kuphuculwe ukuchaneka kokubikezela, njengokhetho). Inqaku eliphambili ekuqondeni indima ye-CT "i-abstraction" kwizinto eziyinyani zeCD kunye neCI: I-CT iya kuphumeza zonke izinto zakudala, igxininise kwiinkcukacha zokubala kunye neemathematika zesisombululo se-AI / ML ngaphakathi kwezakhono ezinikezelwa yimimandla ethile. Uxanduva "lokubonelela ngegalelo" kunye "nokuhambisa iziphumo" luya kuba luxanduva lweCD kunye neCI.

Inkalo ebalulekileyo yokuphunyezwa kwe-CT ngokukodwa kwi-InterSystems IRIS: ukusebenzisa ukudibanisa kunye neendawo zokumisela i-mathematika esele zikhankanywe ngasentla, iqonga linamandla okukhupha ezo zinto zobugcisa kwiiseshoni zokusebenza ezisebenza phantsi kolawulo lwayo kwiimeko zemathematika kwaye (okubaluleke kakhulu) ukujika. zibe zizinto zedatha yeqonga. Ngokomzekelo, itafile yokusabalalisa esandul 'ukudalwa kwiseshoni yePython esebenzayo ingaba (ngaphandle kokumisa iseshoni yePython) idluliselwe kwiqonga ngendlela, umzekelo, ihlabathi (i-multidimensional InterSystems IRIS data array) - kwaye isetyenzisiwe. ukubala kwenye i-AI/ML- indlela (eyenziwe ngolwimi lwenye imeko-bume - umzekelo, ku-R) - okanye itafile yenyani. Omnye umzekelo: ngokuhambelana "nendlela eqhelekileyo" yokusebenza kwemodeli (kwiseshoni yokusebenza yePython), "i-auto-ML" iqhutyelwa kwidatha yayo yokufaka: ukukhethwa ngokuzenzekelayo kwezinto eziguquguqukayo zegalelo kunye namaxabiso eparameter. Kwaye kunye noqeqesho "oluqhelekileyo", imodeli enemveliso ngexesha lokwenyani ikwafumana "isindululo sokuphucula" yenkcazo yayo-apho iseti yezinto eziguquguqukayo zitshintsha, ixabiso leparameter liyatshintsha (akusekho ngenxa yoqeqesho. kwiPython, kodwa ngenxa yoqeqesho kunye "enye" ​​inguqulelo ngokwayo, efana ne-H2O stack), ivumela isisombululo se-AI / ML ngokubanzi ukuba simelane ngokuzimeleyo neenguqu ezingalindelekanga kwindalo yedatha yegalelo kunye neziganeko ezilandelwayo. .

Makhe siqhelane ngakumbi kunye neqonga le-AI / ML yokusebenza kwe-InterSystems IRIS, sisebenzisa umzekelo weprototype yobomi bokwenyani.

Kumzobo ongezantsi, kwicala lasekhohlo lesilayidi kukho inxalenye yenkqubo yoshishino ephunyezwa ukuphunyezwa kwezikripthi kwiPython kunye ne-R. Kwinxalenye ephakathi kukho iilogi ezibonakalayo zokwenziwa kwezinye zezi scripts, ngokulandelanayo, kwiPython kunye ne-R. Ngokukhawuleza emva kwabo yimizekelo yomxholo kwelinye kunye nolunye ulwimi, udluliselwe ukuphunyezwa kwiindawo ezifanelekileyo. Ekugqibeleni ngasekunene yimibono esekelwe kwiziphumo zokwenziwa kwescript. Ukubonwa okuphezulu kwenziwa kwi-IRIS Analytics (idatha ithathwe kwi-Python kwi-InterSystems IRIS iqonga ledatha kwaye iboniswe kwideshibhodi usebenzisa iqonga), ezantsi zenziwe ngokuthe ngqo kwiseshoni yomsebenzi we-R kunye nesiphumo esivela apho kwiifayile zegraphic. . Umba obalulekileyo: iqhekeza elibonisiweyo kwiprototype linoxanduva lokuqeqesha imodeli (ukuhlelwa kweemeko zezixhobo) kwidatha efunyenwe ngexesha lokwenyani ukusuka kwinkqubo yesilingisi sesixhobo, phezu komyalelo wenkqubo yokubeka esweni umgangatho oqatshelwe ngexesha lokusetyenziswa kwemodeli. Ukuphunyezwa kwesisombululo se-AI / ML ngendlela yeesethi zeenkqubo zokusebenzisana ("ii-arhente") ziya kuxoxwa ngakumbi.

Umzobo we-12 Ukusebenzisana kunye nePython, uR kunye noJulia kwi-InterSystems IRIS

Iinkqubo zeqonga (zikwa "ziinkqubo zeshishini", "iinkqubo zohlalutyo", "imibhobho", njl.njl. - ngokuxhomekeke kumxholo), okokuqala, zihlelwa kumhleli wenkqubo yegraphical ishishini kwiqonga ngokwalo, nakwindlela enjalo. indlela yokuba zombini umzobo webhloko kunye ne-AI / ML ehambelanayo indlela (ikhowudi yeprogram) yenziwe ngaxeshanye. Xa sisithi "indlela ye-AI/ML ifunyenwe," ekuqaleni sithetha ukuxutywa (ngaphakathi kwenkqubo enye): umxholo kwiilwimi zeemodelo zemathematika ukufuphi nomxholo kwi-SQL (kubandakanya izandiso ezivela IntegratedML), kwi-InterSystems ObjectScript, kunye nezinye iilwimi ezixhaswayo. Ngaphezu koko, inkqubo yeqonga ibonelela ngamathuba abanzi kakhulu "okunikezela" ngendlela yamaqhekeza abekwe ngokwemigangatho (njengoko kunokubonwa kumzekelo kumzobo ongezantsi), okuvumela ukuba ulungelelanise ngokufanelekileyo umxholo onzima kakhulu ngaphandle kokuba "uphume" yefomathi yemizobo (kwiifomathi “ezingezographical” » iindlela/iiklasi/iinkqubo, njl. njl.). Oko kukuthi, ukuba kukho imfuneko (kwaye kubonwa kwangaphambili kwiiprojekthi ezininzi), ngokupheleleyo isisombululo se-AI/ML sinokuphunyezwa kwifomathi yokuzibhala ngokwegraphical. Nceda uqaphele ukuba kwindawo ephakathi komzobo ongezantsi, omele "inqanaba le-nesting" ephezulu, kuyacaca ukuba ngaphezu komsebenzi wangempela wokuqeqesha imodeli (usebenzisa i-Python kunye ne-R), uhlalutyo lwento ebizwa ngokuba yi-nesting. Ijika le-ROC lemodeli eqeqeshiweyo yongezwa, ivumela ngokubonakalayo (kunye nokubala kwakhona) ukuvavanya umgangatho woqeqesho - kwaye olu hlalutyo luphunyezwa ngolwimi lwaseJulia (lwenziwa, ngokufanelekileyo, kwimeko yezibalo zeJulia).

Umzobo 13 Ubume obubonakalayo bokuqulunqwa kwezisombululo ze-AI/ML kwi-InterSystems IRIS

Njengoko kukhankanyiwe ngaphambili, uphuhliso lokuqala kunye (kwezinye iimeko) ukulungelelaniswa kweendlela ze-AI / ML esele ziphunyeziwe kwiqonga liya kwenziwa ngaphandle kweqonga kumhleli weJupyter. Kumzobo ongezantsi sibona umzekelo wokulungelelanisa inkqubo yeqonga ekhoyo (efanayo nakulo mfanekiso ungentla) - yile ndlela iqhekeza elijongene nokuqeqesha imodeli libukeka ngayo kwiJupyter. Umxholo wePython uyafumaneka ukuze uhlele, ulungiswe, kunye nemveliso yegraphics ngqo kwiJupyter. Utshintsho (ukuba luyimfuneko) lunokwenziwa ngolungelelwaniso olukhawulezileyo kwinkqubo yeqonga, kubandakanywa nohlobo lwayo oluvelisayo. Umxholo omtsha unokudluliselwa kwiqonga ngendlela efanayo (inkqubo entsha yeqonga yenziwa ngokuzenzekelayo).

Umzobo we-14 usebenzisa iJupyter Notebook ukuhlela injini ye-AI/ML kwiqonga leInterSystems IRIS

Uhlengahlengiso lwenkqubo yeqonga lungenziwa kuphela kwifomathi yegraphical okanye yelaptop - kodwa nakwifomathi “yetotali” ye-IDE (Imo yoPhuhliso eDityanisiweyo). Ezi IDE ziyi-IRIS Studio (i-studio ye-IRIS yendabuko), iKhowudi ye-Visual Studio (I-InterSystems IRIS extension ye-VSCode) kunye ne-Eclipse (i-plugin ye-Atelier). Kwezinye iimeko, kunokwenzeka ukuba iqela lophuhliso lisebenzise zonke ii-IDE ezintathu ngaxeshanye. Umzobo ongezantsi ubonisa umzekelo wokuhlela inkqubo efanayo kwi-studio ye-IRIS, kwiKhowudi ye-Visual Studio kunye ne-Eclipse. Ngokuqinisekileyo yonke imixholo iyafumaneka ukuze ihlelwe: Python/R/Julia/SQL, ObjectScript, kunye nenkqubo yeshishini.

Umzobo we-15 Uphuhliso lwenkqubo ye-InterSystems ye-IRIS kwii-IDE ezahlukeneyo

Izixhobo zokuchaza kunye nokwenza iinkqubo ze-InterSystems ze-IRIS kwiiLwimi zeNkqubo yoShishino (BPL) zifanelwe ukukhankanywa ngokukodwa. I-BPL yenza kube lula ukusebenzisa "amacandelo okudibanisa esele enziwe" (imisebenzi) kwiinkqubo zoshishino - ngokwenene, inika isizathu sokuba "ukuhlanganiswa okuqhubekayo" kuphunyezwa kwi-InterSystems IRIS. Amacandelo enkqubo yoshishino esele yenziwe (imisebenzi kunye nokudibanisa phakathi kwabo) i-accelerator enamandla yokuhlanganisa isisombululo se-AI / ML. Kwaye kungekhona kuphela iindibano: ngenxa yemisebenzi kunye nokudibanisa phakathi kwabo ngaphezu kophuhliso oluhlukeneyo lwe-AI / ML kunye neendlela, "uluhlu lolawulo oluzimeleyo" luvela, olukwazi ukwenza izigqibo ngokwemeko, ngexesha langempela.

Umzobo we-16 Amacandelo enkqubo yoshishino esele ilungele ukudityaniswa ngokuqhubekayo (CI) kwiqonga le-InterSystems IRIS

Ingcamango yeenkqubo ze-agent (eyaziwa nangokuthi "iinkqubo ze-agent ezininzi") inesimo esinamandla kwi-robotization, kunye ne-InterSystems IRIS iqonga le-organically lixhasa nge-"product-product" yokwakhiwa. Ukongeza kumathuba angenamkhawulo "wokufaka" inkqubo nganye kunye nokusebenza okuyimfuneko kwisisombululo esipheleleyo, ukunika inkqubo yenkqubo yeqonga kunye nepropathi ye "arhente" ikuvumela ukuba wenze izisombululo ezisebenzayo kwizinto ezizinzile kakhulu ezifanisiweyo (ukuziphatha kwentlalo / i-biosystems, iinkqubo zobuchwepheshe ezinokubonwa ngokuyinxenye, njl. njl.).

Umzobo we-16 Ukusebenza kwesisombululo se-AI/ML njengenkqubo yenkqubo yoshishino esekwe kwi-arhente kwi-InterSystems IRIS

Siqhubeka nophononongo lwethu lwe-InterSystems IRIS ngebali malunga nokusetyenziswa kweqonga lokusombulula iiklasi ezipheleleyo zeengxaki zexesha langempela (intshayelelo eneenkcukacha ezichanekileyo kwezinye zeendlela ezilungileyo zeqonga i-AI/ML kwi-InterSystems IRIS inokufumaneka kwenye yethu yangaphambili iiwebinars).

Kushushu kwizithende zomzobo wangaphambili, ngezantsi kukho umzobo ocacileyo wenkqubo ye-ejenti. Umzobo ubonisa iprototype efanayo, zonke iinkqubo ezine ze-ejenti ziyabonakala, ubudlelwane phakathi kwabo buzotywe ngokucwangcisiweyo: I-GENERATOR - inkqubo yokwenziwa kwedatha ngabenzi boluvo lwezixhobo, i-BUFFER - ilawula imigca yedatha, i-ANALYZER - yenza umatshini wokufunda ngokwawo, UMONI - ubeka iliso umgangatho wokufunda koomatshini kwaye ungenise uphawu malunga nesidingo sokuqeqesha kwakhona imodeli.

Umzobo we-17 Ukuqulunqwa kwesisombululo se-AI / ML ngendlela yenkqubo yenkqubo ye-arhente esekelwe kwi-InterSystems IRIS

Lo mzobo ungezantsi ubonisa ukusebenza ngokuzimeleyo kwenye iprototype yerobhothi (ukuqatshelwa kombala weemvakalelo kwizicatshulwa) kangangexesha elithile. Kwinxalenye ephezulu yinguquko yesalathisi somgangatho woqeqesho lwemodeli (umgangatho ukhula), kwinqanaba elisezantsi liguquguqukayo lwesalathisi somgangatho wesicelo somzekelo kunye neenyaniso zoqeqesho oluphindaphindiweyo (imivimbo ebomvu). Njengoko ubona, isisombululo sifunde ngokufanelekileyo nangokuzimela, kwaye sisebenza kwinqanaba elinikiweyo lomgangatho (amaxabiso asemgangathweni aweli ngaphantsi kwe-80%).

Umzobo we-18 oqhubekayo (self-) uqeqesho (CT) kwi-InterSystems IRIS iqonga

Siphinde sakhankanya "i-auto-ML" ngaphambili, kodwa umzobo ongezantsi ubonisa ukusetyenziswa kwalo msebenzi ngokweenkcukacha usebenzisa umzekelo wenye iprototype. Umzobo womzobo weqhekeza lenkqubo yeshishini ubonisa umsebenzi obangela umfuziselo kwisitaki se-H2O, ubonisa iziphumo zolu mfuziselo (ulawulo olucacileyo lwemodeli eyisiphumo kwiimodeli “ezenziwe ngabantu”, ngokomzobo wokuthelekisa I-ROC curves, kunye nokuchongwa ngokuzenzekelayo "kwezona ziguquko zinempembelelo" ezikhoyo kwisethi yedatha yokuqala). Inqaku elibalulekileyo apha kukugcinwa kwexesha kunye nezibonelelo zeengcali eziphunyezwa nge "auto-ML": yintoni inkqubo yethu yeqonga eyenziwa kwisiqingatha somzuzu (ukufumana kunye nokuqeqesha imodeli efanelekileyo) inokuthatha ingcali ukusuka kwiveki ukuya kwinyanga.

Umzobo we-19 Ukuhlanganiswa kwe "auto-ML" kwisisombululo se-AI / ML kwi-InterSystems IRIS iqonga

Umzobo ongezantsi ulahlekelwa yinto encinci, kodwa yindlela efanelekileyo yokuphelisa ibali malunga neeklasi zeengxaki zexesha langempela ezisombululwa: siyakukhumbuza ukuba kunye nawo onke amandla e-InterSystems IRIS platform, imodeli yoqeqesho phantsi kolawulo lwayo. hayi isinyanzelo. Iqonga linokufumana ngaphandle kwento ebizwa ngokuba yi-PMML yokucaciswa kwemodeli, eqeqeshwe kwisixhobo esingekho phantsi kolawulo lweqonga - kwaye isebenzise le modeli ngexesha lokwenyani ukusuka ekungenisweni kwayo. Iimpawu zePMML. Kubalulekile ukuqwalasela ukuba ayizizo zonke izinto ze-AI / ML ezinokuncitshiswa kwinkcazo ye-PMML, nokuba uninzi lwezona zinto ziqhelekileyo zivumela oku. Ngaloo ndlela, iqonga le-InterSystems IRIS "i-loop evulekile" kwaye ayithethi "ubukhoboka beqonga" kubasebenzisi.

Umzobo we-20 Ukuhlanganiswa kwe "auto-ML" kwisisombululo se-AI / ML kwi-InterSystems IRIS iqonga

Makhe sidwelise iingenelo zeqonga ezongezelelweyo ze-InterSystems IRIS (ukucaca, ngokubhekiselele kulawulo lwenkqubo), ezibaluleke kakhulu kwi-automation ye-intelligence intelligence kunye nokufunda komatshini wangempela:

• Ukuphuhliswa kwezixhobo zokudibanisa nayo nayiphi na imithombo yedatha kunye nabathengi (inkqubo yokulawula inkqubo / i-SCADA, izixhobo, i-MRO, i-ERP, njl.)
• Yakhelwe ngaphakathi Iimodeli ezininzi ze-DBMS ukusebenza okuphezulu kwetransaction kunye nohlalutyo (i-Hybrid Transaction / Analytical Processing, HTAP) yayo nayiphi na umthamo wedatha yenkqubo yeteknoloji
• Izixhobo zokuphuhlisa ukuhanjiswa okuqhubekayo kweenjini ze-AI/ML kwizisombululo zexesha langempela ezisekelwe kwiPython, R, Julia.
• Iinkqubo zoshishino ezilungelelanisiweyo zokudityaniswa okuqhubekayo kunye (nokwakho) ukufunda ngexesha langempela leenjini zesisombululo ze-AI/ML
• Izixhobo ezakhelwe kwiBusiness Intelligence yokujonga idatha yenkqubo kunye neziphumo zesisombululo se-AI / ML
• Ulawulo lwe-API ukuhambisa iziphumo zesisombululo se-AI / ML kwinkqubo yokulawula iinkqubo / i-SCADA, ulwazi kunye neenkqubo zokuhlalutya, ukuthumela izilumkiso, njl.

Izisombululo ze-AI / ML kwi-InterSystems IRIS iqonga lingena lula kwisiseko se-IT esikhoyo. Iqonga le-InterSystems IRIS liqinisekisa ukuthembeka okuphezulu kwezisombululo ze-AI / ML ngokuxhasa ukucwangciswa kokunyamezela kunye nokunyamezela kweentlekele kunye nokuhanjiswa okuguquguqukayo kwiindawo ezibonakalayo, kwiiseva ezibonakalayo, kumafu abucala kunye noluntu, kunye nezikhongozeli zeDocker.

Ngaloo ndlela, i-InterSystems IRIS yinkqubo yekhompyutha ye-AI / ML yexesha langempela. I-universality yeqonga lethu liqinisekiswa ekusebenzeni ngokungabikho kwezithintelo ze-de facto kubunzima bokubala okuphunyeziweyo, ukukwazi kwe-InterSystems IRIS ukudibanisa (ngexesha langempela) ukuqhutyelwa kweemeko ezivela kwiindidi ezahlukeneyo zamashishini, kunye nokulungelelaniswa okukhethekileyo. nayiphi na imisebenzi yeqonga kunye neendlela kwiimfuno ezithile zabasebenzisi.

Umzobo we-21 InterSystems IRIS - iqonga lekhompyuter le-AI / ML yendalo yonke

Ngonxibelelwano olubambekayo kunye nabafundi bethu abanomdla kumbandela oboniswe apha, sicebisa ukuba ungazibekeli kumda wokuyifunda kwaye uqhubeke nencoko "bukhoma." Siya kukuvuyela ukubonelela ngenkxaso ngokuqulunqwa kweemeko ze-AI / ML ngexesha langempela ngokumalunga neenkcukacha zenkampani yakho, yenza i-prototyping edibeneyo kwiqonga le-InterSystems IRIS, uqulunqe kwaye uphumeze ngokusebenza imephu yendlela yokwazisa ubukrelekrele bokwenziwa kunye nokufunda komatshini. kwimveliso yakho kunye neenkqubo zolawulo. I-imeyile yethu yoQhagamshelwano leNgcali ye-AI/ML – [imeyile ikhuselwe].

umthombo: www.habr.com