I-InterSystems IRIS - inkundla ye-AI/ML yesikhathi sangempela yendawo yonke

Umbhali: Sergey Lukyanchikov, unjiniyela onguchwepheshe kwa-InterSystems

Izingcingo zekhompuyutha zesikhathi sangempela ze-AI/ML

Ake siqale ngezibonelo ezivela kokuhlangenwe nakho kwe-Data Science practice kwa-InterSystems:

• Iphothali yomthengi elayishiwe ixhunywe kusistimu yezincomo eziku-inthanethi. Kuzoba nokuhlelwa kabusha kwamaphromoshini kuyo yonke inethiwekhi yokudayisa (isibonelo, esikhundleni solayini “oyisicaba” wamaphromoshini, kuzosetshenziswa i-matrix “segment-tactics”). Kwenzekani ezinjinini zokuncoma? Kwenzekani ekuhanjisweni nasekubuyekezweni kwedatha enjinini yokuncoma (ivolumu yedatha yokufaka inyuke izikhathi ezingu-25000)? Kwenzekani ekuthuthukisweni kwezincomo (isidingo sokunciphisa umkhawulo wokuhlunga wemithetho yokuncoma ngokuphindwe kayinkulungwane ngenxa yokwanda okuphindwe inkulungwane kwenani lazo kanye “nobubanzi”)?
• Kunohlelo lokuqapha amathuba okuba kube khona izinkinga ezikhulayo ezingxenyeni zemishini. Isistimu yokulawula inqubo ezenzakalelayo yayixhunywe ohlelweni lokuqapha, idlulisa izinkulungwane zamapharamitha wenqubo yezobuchwepheshe njalo ngomzuzwana. Kwenzekani ohlelweni lokuqapha olwake lwasebenza "kumasampula okwenziwa ngesandla" (ingabe luyakwazi ukuhlinzeka ngokuqapha okungenzeka kwesibili nesekhondi)? Kuzokwenzekani uma ibhulokhi entsha yamakholomu angamakhulu amaningana ibonakala kudatha yokufaka nokufundwa kwezinzwa ezisanda kufakwa ohlelweni lokulawula inqubo (kuzodingeka futhi isikhathi esingakanani ukumisa uhlelo lokuqapha ukufaka idatha evela kuzinzwa ezintsha ekuhlaziyeni )?
• Isethi yezindlela ze-AI/ML (izincomo, ukuqapha, ukubikezela) idaliwe ezisebenzisa imiphumela yomsebenzi womunye nomunye. Mangaki amahora omuntu adingekayo inyanga ngayinye ukulungisa ukusebenza kwale nkimbinkimbi nezinguquko kudatha yokufaka? Kuyini “ukwehla kwejubane” okuvamile uma kusekelwa inkimbinkimbi yokwenza izinqumo zabaphathi (imvamisa yokwenzeka kolwazi olusha olusekelayo kuyo uma kuqhathaniswa nemvamisa yokwenzeka kwedatha entsha yokufaka)?

Uma sifingqa lezi nezinye izibonelo eziningi, sifike ekwakhiweni kwezinselelo eziphakamayo lapho sithuthela ekusetshenzisweni komshini wokufunda nezindlela zobuhlakani bokwenziwa ngesikhathi sangempela:

• Ingabe sanelisekile ngesivinini sokudala nokuzivumelanisa nezimo (esimeni esishintshayo) sentuthuko ye-AI/ML enkampanini yethu?
• Ingabe izixazululo ze-AI/ML esizisebenzisayo zisekela kangakanani ukuphathwa kwebhizinisi kwesikhathi sangempela?
• Ingabe izixazululo ze-AI/ML esizisebenzisayo ziyakwazi ukuzivumelanisa noshintsho kudatha nezinqubo zokuphatha ibhizinisi ngokuzimela (ngaphandle konjiniyela)?

I-athikili yethu ingumbono ophelele wamakhono enkundla ye-InterSystems IRIS mayelana nokusekelwa kwendawo yonke yokuthunyelwa kwezindlela ze-AI/ML, ukuhlanganisa (ukuhlanganiswa) kwezixazululo ze-AI/ML, nokuqeqeshwa (ukuhlolwa) kwezixazululo ze-AI/ML ngokujulile. ukusakazwa kwedatha. Sizobheka ucwaningo lwemakethe, izibonelo zezixazululo ze-AI/ML, nezici zomqondo zalokho esikubiza ngokuthi inkundla yesikhathi sangempela ye-AI/ML kulesi sihloko.

Esikwaziyo kuzinhlolovo: izinhlelo zokusebenza zesikhathi sangempela

Imiphumela inhlolovoeyenziwe phakathi kochwepheshe be-IT abacishe babe ngu-800 ngo-2019 yiLightbend, bazikhulumele:

Umfanekiso 1 Abathengi abahamba phambili bedatha yesikhathi sangempela

Ake sicaphune izingcezu ezibalulekile zombiko ngemiphumela yalolu cwaningo ekuhumusheni kwethu:

"... Amathrendi ekudumeni kwamathuluzi okuhlanganisa ukusakazwa kwedatha futhi, ngesikhathi esifanayo, ukusekela ikhompuyutha ezitsheni kunikeza impendulo evumelanayo esicelweni semakethe sokuphendula okwengeziwe, okunengqondo, isiphakamiso esiguquguqukayo sezixazululo ezisebenzayo. Ukusakaza-bukhoma idatha kudlulisa ulwazi ngokushesha kunedatha yephakethe yendabuko. Okungezwe kulokhu yikhono lokusebenzisa ngokushesha izindlela zokubala, ezifana, isibonelo, izincomo ezisekelwe ku-AI/ML, ukudala izinzuzo zokuncintisana ngokukhuphuka kokwaneliseka kwamakhasimende. Umjaho we-agility uphinda ube nomthelela kuzo zonke izindima ku-paradigm ye-DevOps - okwenza ukuthuthukiswa kwezinhlelo zokusebenza nokusatshalaliswa kusebenze kahle kakhulu. … Ochwepheshe be-IT abangamakhulu ayisishiyagalombili nane banikeze ulwazi ngokusetshenziswa kokugeleza kwedatha ezinhlanganweni zabo. Abaphendulile babetholakala kakhulu emazweni aseNtshonalanga (ama-41% e-Europe kanye nama-37% eNyakatho Melika) futhi bacishe basakazwa ngokulinganayo phakathi kwezinkampani ezincane, eziphakathi nezinkulu. ...

... Ubuhlakani bokwenziwa akuwona ama-hype. Amaphesenti angamashumi amahlanu nesishiyagalombili alabo asebesebenzisa ukucutshungulwa kokusakazwa kwedatha ekukhiqizeni izinhlelo zokusebenza ze-AI/ML baqinisekisa ukuthi ukusebenzisa kwabo i-AI/ML kuzobona ukukhula okukhulu kakhulu onyakeni olandelayo (uma kuqhathaniswa nezinye izinhlelo zokusebenza).

• Ngokusho kweningi labaphendulile, ukusetshenziswa kokusakaza kwedatha kuzimo ze-AI/ML kuzobona ukukhula okukhulu kakhulu onyakeni olandelayo.
• Izinhlelo zokusebenza ku-AI/ML zizokhula hhayi kuphela ngenxa yezinhlobo ezintsha zezimo, kodwa futhi ngenxa yezimo ezivamile lapho idatha yesikhathi sangempela isetshenziswa kakhulu.
• Ngaphezu kwe-AI/ML, izinga lentshiseko phakathi kwabasebenzisi bamapayipi edatha ye-IoT liyamangaza - ama-48% alabo asebevele bahlanganise idatha ye-IoT bathi ukuqaliswa kwesimo kule datha kuzobona ukwanda okukhulu esikhathini esizayo esiseduze. ..."

Kulolu cwaningo oluthakazelisa kakhulu, kuyacaca ukuthi umbono wokufunda ngomshini kanye nezimo zobuhlakani bokwenziwa njengabaholi ekusetshenzisweni kwemifudlana yedatha kakade "usendleleni." Kodwa ukuqaphela okubaluleke ngokufanayo umbono wesikhathi sangempela se-AI/ML ngokusebenzisa ilensi ye-DevOps: lapha sesingaqala ukukhuluma ngoguquko lwesiko elisabusa kakhulu “le-AI/ML elahlwayo ngesethi yedatha efinyeleleka ngokugcwele.”

Umqondo weplathifomu yesikhathi sangempela ye-AI/ML

Indawo eyodwa evamile yesicelo se-AI/ML yesikhathi sangempela ukulawulwa kwenqubo ekukhiqizeni. Sisebenzisa isibonelo sakhe futhi sicabangela imicabango yangaphambilini, sizokwakha umqondo wenkundla yesikhathi sangempela ye-AI/ML.
Ukusetshenziswa kobuhlakani bokwenziwa nokufunda komshini ekulawuleni inqubo kunezici ezimbalwa:

• Idatha yesimo senqubo yezobuchwepheshe yamukelwe kakhulu: ngemvamisa ephezulu kanye nohlu olubanzi lwamapharamitha (kufika emashumini ezinkulungwane zamanani wepharamitha adluliselwa ngomzuzwana kusuka ohlelweni lokulawula inqubo)
• Idatha ekuhlonzeni amaphutha, ingasaphathwa ngedatha ekuthuthukisweni kwayo, ngokuphambene nalokho, iyindlala futhi ayijwayelekile, ebonakala ngokungafaneli okwanele kweziphambeko kanye nokwenza kwasendaweni ngesikhathi (imvamisa imelelwa amarekhodi ephepha)
• Ngokombono ongokoqobo, kuphela “iwindi lokuhambisana” ledatha yomthombo elitholakalayo ukuze kuqeqeshwe futhi kusetshenziswe amamodeli, abonisa amandla enqubo yezobuchwepheshe phakathi nesikhathi sokuslayida esinengqondo esiphetha ngamavelu okugcina okufundwa kwamapharamitha wenqubo.

Lezi zici ziyasiphoqa, ngaphezu kokwamukela kanye nokucutshungulwa okuyisisekelo ngesikhathi sangempela "sesiginali yokufaka i-broadband" ejulile kusukela kunqubo yezobuchwepheshe, ukwenza (ngokufanayo) isicelo, ukuqeqeshwa kanye nokulawulwa kwekhwalithi yemiphumela ye-AI / Amamodeli we-ML - futhi ngesikhathi sangempela. “Uhlaka” amamodeli ethu “alubonayo” efasiteleni elislayidayo lokufaneleka luhlala lushintsha - futhi ngalo, ikhwalithi yemiphumela yomsebenzi wamamodeli e-AI/ML aqeqeshwe kwenye “yezinhlaka” esikhathini esidlule nayo iyashintsha. . Uma izinga lemiphumela yomsebenzi wamamodeli e-AI/ML liwohloka (isibonelo: inani lephutha lesigaba “le-alamu-norm” leqe umkhawulo esiwuchazile), ukuqeqeshwa okwengeziwe kwamamodeli kufanele kuqaliswe ngokuzenzakalelayo "uhlaka" lwamanje - kanye nokukhetha isikhathi sokuqalisa ukuqeqeshwa okwengeziwe kwamamodeli kufanele kubhekwe ukuthi isikhathi sokuqeqeshwa ngokwaso, kanye namandla okuwohloka kwekhwalithi yomsebenzi wenguqulo yamanje yamamodeli (njengoba izinguqulo zamanje zamamodeli ziyaqhubeka nokusetshenziswa ngenkathi amamodeli eqeqeshwa, futhi kuze kube yilapho kwakhiwa izinguqulo zabo "ezisanda kuqeqeshwa").

I-InterSystems IRIS inamandla epulatifomu abalulekile okunika amandla izixazululo ze-AI/ML zokulawula inqubo yesikhathi sangempela. Lawa makhono angahlukaniswa ngamaqembu amathathu amakhulu:

• Ukuthunyelwa okuqhubekayo (Ukuthunyelwa Okuqhubekayo/Ukulethwa, i-CD) yezindlela ezintsha noma ezishintshiwe ezikhona ze-AI/ML zibe isixazululo esikhiqizayo esisebenza ngesikhathi sangempela kuplathifomu ye-InterSystems IRIS
• Ukuhlanganiswa Okuqhubekayo (CI) kusixazululo esisodwa esikhiqizayo sokugeleza kwedatha yenqubo yezobuchwepheshe engenayo, imigqa yedatha yokusetshenziswa/ukuqeqeshwa/ukulawulwa kwekhwalithi yezinqubo ze-AI/ML kanye nokushintshisana kwedatha/izenzo zekhodi/zokulawula ngendawo yokumodela yezibalo, ehlelwe ngesikhathi sangempela. Isiteji se-InterSystems IRIS
• Ukuqeqeshwa okuqhubekayo (self-) (Ukuqeqeshwa Okuqhubekayo, CT) kwezindlela ze-AI/ML, okwenziwa ezindaweni zokumodela zezibalo kusetshenziswa idatha, ikhodi kanye nezenzo zokulawula ("izinqumo ezenziwe") ezidluliselwa yiplathifomu ye-InterSystems IRIS

Ukuhlukaniswa kwamakhono enkundla ngokuhlobene nokufunda komshini nobuhlakani bokwenziwa kulawo maqembu akwenzeki ngephutha. Ake sicaphune indlela yokwenza ukushicilelwa I-Google, ehlinzeka ngesisekelo somqondo salokhu kuhlukaniswa, ekuhumusheni kwethu:

“... Umqondo we-DevOps, odumile kulezi zinsuku, uhlanganisa ukuthuthukiswa nokusebenza kwezinhlelo zolwazi ezinkulu. Izinzuzo zokusebenzisa lo mqondo ukuncishiswa kobude besikhathi somjikelezo wokuthuthukisa, ukuthunyelwa okusheshayo kwentuthuko, kanye nokuguquguquka ekuhleleni ukukhishwa. Ukuze kuzuzwe lezi zinzuzo, i-DevOps ibandakanya ukusetshenziswa okungenani kwezinqubo ezimbili:

• Ukuhlanganiswa Okuqhubekayo (CI)
• Ukulethwa Okuqhubekayo (CD)

Lezi zinqubo zisebenza nasezinkundleni ze-AI/ML ukuze kuqinisekiswe ukuhlangana okuthembekile nokusebenzayo kwezixazululo ezikhiqizayo ze-AI/ML.

Izinkundla ze-AI/ML ziyahluka kwezinye izinhlelo zolwazi kulezi zici ezilandelayo:

• Amakhono eqembu: Lapho udala isisombululo se-AI/ML, ithimba ngokuvamile lihlanganisa ososayensi bedatha noma ochwepheshe “bezemfundo” emkhakheni wocwaningo lwedatha abenza ukuhlaziya idatha, bathuthukise futhi bahlole amamodeli. Lawa malungu eqembu angahle angabi onjiniyela bekhodi abakhiqizayo.
• Ukuthuthukiswa: Izinjini ze-AI/ML ziwukulinga ngokwemvelo. Ukuze uxazulule inkinga ngendlela ephumelela kakhulu, kuyadingeka ukuthi udlule ezinhlanganisela ezihlukahlukene zokufakwayo, ama-algorithms, izindlela zokumodela kanye nemingcele yamamodeli. Ubunkimbinkimbi bosesho olunjalo busekulandeleni "okusebenzile/okungazange kusebenze", ukuqinisekisa ukuphindaphindeka kweziqephu, ukwenziwa okuvamile kwentuthuko ukuze kusetshenziswe ngokuphindaphindiwe.
• Ukuhlola: Ukuhlola izinjini ze-AI/ML kudinga uhla olubanzi lokuhlola kunezinye izinto eziningi ezithuthukayo. Ngaphezu kokuhlolwa kweyunithi evamile nokuhlanganiswa, ukufaneleka kwedatha kanye nekhwalithi yemiphumela yokusebenzisa imodeli ekuqeqesheni nasekulawuleni amasampula kuyahlolwa.
• Ukuthunyelwa: Ukuthunyelwa kwezixazululo ze-AI/ML akukhawulelwe kumasevisi aqagelayo asebenzisa imodeli eke yaqeqeshwa. Izixazululo ze-AI/ML zakhiwe eduze kwamapayipi ezigaba eziningi enza ukuqeqeshwa okuzenzakalelayo kwemodeli nokusebenzisa. Ukukhipha amaphayiphi anjalo kuhilela ukuzenzela izenzo ezingezona ezincane ngokuvamile ezenziwa ngesandla ngososayensi bedatha ukuze bakwazi ukuqeqesha nokuhlola amamodeli.
• Ukukhiqiza: Izinjini ze-AI/ML azikwazi ukuntula ukukhiqiza hhayi nje ngenxa yokungasebenzi kahle kwezinhlelo, kodwa nangenxa yokushintshashintsha kwemvelo yedatha yokufaka. Ngamanye amazwi, ukusebenza kwezinqubo ze-AI/ML kungehlisa isithunzi ngenxa yohlu olubanzi lwezizathu kunokusebenza kwentuthuko evamile. Okuholela esidingweni sokuqapha (ku-inthanethi) ukusebenza kwezinjini zethu ze-AI/ML, kanye nokuthumela izexwayiso noma ukwenqaba imiphumela uma izinkomba zokusebenza zingahlangabezani nokulindelwe.

Izinkundla ze-AI/ML zifana nezinye izinhlelo zolwazi ngokuthi zombili zidinga ukuhlanganiswa kwekhodi okuqhubekayo nokulawulwa kwenguqulo, ukuhlolwa kweyunithi, ukuhlolwa kokuhlanganiswa, kanye nokuthunyelwa kwentuthuko okuqhubekayo. Kodwa-ke, endabeni ye-AI/ML, kunomehluko eminingana obalulekile:

• I-CI (Ukuhlanganiswa Okuqhubekayo) ayisalinganiselwe ekuhloleni nasekuqinisekiseni ikhodi yezingxenye ezisetshenzisiwe - futhi ihlanganisa ukuhlola nokuqinisekisa idatha kanye namamodeli e-AI/ML.
• I-CD (Ukulethwa Okuqhubekayo/Ukuthunyelwa, ukuthunyelwa okuqhubekayo) ayikhawulelwe ekubhaleni nasekukhipheni amaphakheji noma izinsiza, kodwa isho inkundla yokubunjwa, ukuqeqeshwa nokusebenzisa izixazululo ze-AI/ML.
• I-CT (Ukuqeqeshwa Okuqhubekayo, ukuqeqeshwa okuqhubekayo) isici esisha [cishe. umbhali we-athikili: into entsha ngokuhlobene nomqondo wendabuko we-DevOps, lapho i-CT, njengomthetho, Ukuhlola Okuqhubekayo], etholakala kumapulatifomu e-AI/ML, ebhekele ukuphathwa okuzimele kwezindlela zokuqeqesha nokusebenzisa i-AI. /ML amamodeli. ..."

Singasho ukuthi ukufunda ngomshini nobuhlakani bokwenziwa obusebenza kudatha yesikhathi sangempela kudinga isethi ebanzi yamathuluzi nekhono (kusuka ekuthuthukisweni kwekhodi ukuya ekuhleleni izindawo zokumodela zezibalo), ukuhlanganiswa okuseduze phakathi kwazo zonke izindawo zokusebenza nezihloko, ukuhlelwa okusebenza kahle kakhulu komuntu izinsiza zomshini.

Isimo sesikhathi sangempela: ukuqaphela ukuthuthukiswa kokukhubazeka kumaphampu okuphakelayo

Ukuqhubeka nokusebenzisa indawo yokulawula inqubo njengesibonelo, cabanga ngenkinga ethile (esivele siyishilo ekuqaleni): sidinga ukuhlinzeka ngokuqapha kwesikhathi sangempela sokuthuthukiswa kwamaphutha kumaphampu ngokususelwa ekugelezeni kwamanani epharamitha yenqubo. kanye nemibiko evela kubasebenzi abalungisayo mayelana neziphambeko ezihlonziwe.

Umfanekiso 2 Ukwakhiwa kwenkinga yokuqapha ukuthuthukiswa kokukhubazeka

Isici semisebenzi eminingi evezwa ngale ndlela ekusebenzeni ukuthi ukujwayela nokusebenza kahle kokutholwa kwedatha (i-APCS) kufanele kucatshangelwe ngokwasemuva kwesenzakalo sesiqephu nokungavamile (nokubhaliswa) kokukhubazeka kwezinhlobo ezihlukahlukene. Ngamanye amazwi: idatha evela ohlelweni lokulawula inqubo ifika kanye ngomzuzwana, ilungile futhi inembile, futhi amanothi mayelana nokukhubazeka enziwa ngepensela yekhemikhali ekhombisa usuku encwadini yokubhalela ejwayelekile endaweni yokusebenzela (isibonelo: “12.01 – kuvuza kukhava kusukela ohlangothini lwe-3rd bear ").

Ngakho-ke, singakwazi ukwengeza ukwakhiwa kwenkinga ngomkhawulo olandelayo obalulekile: "sinelebula" eyodwa kuphela yokukhubazeka kohlobo oluthile (okungukuthi, isibonelo sokukhubazeka kohlobo oluthile simelelwa idatha evela ekulawuleni inqubo. uhlelo ngosuku oluthile - futhi asinazo izibonelo ezengeziwe zokukhubazeka kwalolu hlobo oluthile). Lo mkhawulo ngokushesha usiyisa ngaphezu kobubanzi bokufunda komshini wakudala (ukufunda okugadiwe), okufanele kube khona "amathegi" amaningi.

Umfanekiso 3 Ukucaciswa komsebenzi wokuqapha ukuthuthukiswa kokukhubazeka

Ingabe ngandlela thile singakwazi “ukwandisa” okuwukuphela “umaka” esinawo? Yebo singakwazi. Isimo samanje sepompo sibonakala ngezinga lokufana neziphambeko ezibhalisiwe. Ngisho nangaphandle kokusetshenziswa kwezindlela zokulinganisa, ezingeni lokubona okubukwayo, ngokubheka ukuguquguquka kwamanani wedatha avela ohlelweni lokulawula inqubo, usungafunda okuningi:

Umfanekiso 4 Amandla wesimo sepompo ngokumelene nengemuva "lophawu" lokukhubazeka kohlobo oluthile.

Kodwa umbono obonakalayo (okungenani okwamanje) akuyona ijeneretha efanelekile "yamathegi" esimweni sethu esishintsha ngokushesha. Sizohlola ukufana kwesimo sepompo samanje neziphambeko ezibikiwe sisebenzisa ukuhlolwa kwezibalo.

Umfanekiso 5 Ukusebenzisa ukuhlolwa kwezibalo kudatha engenayo ngokumelene nesizinda sesici "ilebula"

Uhlolo lwezibalo lunquma amathuba okuthi amarekhodi anamanani amapharamitha wenqubo yobuchwepheshe “kuphakethe lokugeleza” atholwe ohlelweni lokulawula inqubo afana namarekhodi “kamaka” wesici sohlobo oluthile. Inani lamathuba (inkomba yokufana kwezibalo) elibalwe njengomphumela wokusebenzisa uhlolo lwezibalo liguqulwa libe inani elingu-0 noma u-1, libe “ilebula” lokufunda komshini kurekhodi ngalinye elihlolelwa ukufana. Okusho ukuthi, ngemva kokucubungula iphakheji elisha elitholiwe lamarekhodi esimo sepompo ngokuhlolwa kwezibalo, sinethuba (a) lokwengeza le phakheji kusethi yokuqeqesha yokuqeqesha imodeli ye-AI/ML kanye (b) nokulawula ikhwalithi inguqulo yamanje yemodeli uma uyisebenzisa kule phakheji.

Umfanekiso 6 Ukusebenzisa imodeli yokufunda yomshini kudatha engenayo ngokumelene nesizinda "lebula" elinesici.

Kokunye kwethu kwangaphambilini ama-webinars Sikhombisa futhi sichaze ukuthi iplatifomu ye-InterSystems IRIS ikuvumela kanjani ukuthi usebenzise noma iyiphi indlela ye-AI/ML ngendlela yokusebenzisa ngokuqhubekayo izinqubo zebhizinisi eziqapha ukwethembeka kwemiphumela yokumodela futhi uvumelanise amapharamitha wamamodeli. Lapho sisebenzisa i-prototype yesimo sethu ngamaphampu, sisebenzisa konke ukusebenza kwe-InterSystems IRIS okwethulwa phakathi ne-webinar - sisebenzisa inqubo yokuhlaziya njengengxenye yesixazululo sethu hhayi ukufunda okugadiwe okujwayelekile, kodwa kunalokho ukufunda okuqiniswayo, okuphatha ngokuzenzakalela ukukhethwa kwamamodeli okuqeqesha. . Isampula lokuqeqeshwa liqukethe amarekhodi lapho “imvumelwano yokuthola” ivela khona ngemva kokusebenzisa kokubili ukuhlolwa kwezibalo kanye nenguqulo yamanje yemodeli - okungukuthi, kokubili ukuhlolwa kwezibalo (ngemuva kokuguqula inkomba yokufana ibe ngu-0 noma 1) kanye nemodeli ikhiqize umphumela. kulawo marekhodi 1. Phakathi nokuqeqeshwa okusha kwemodeli, ngesikhathi sokuqinisekiswa kwayo (imodeli esanda kuqeqeshwa isetshenziswa kusampula yayo yokuqeqeshwa, kanye nokusetshenziswa kokuqala kokuhlolwa kwezibalo kuyo), amarekhodi “angazange agcine” umphumela woku-1 ngemva kokucubungula. ngokuhlolwa kwezibalo (ngenxa yokuba khona njalo ekuqeqeshweni isampula yamarekhodi asuka “kulebula” yoqobo yesici), ayakhishwa kusethi yokuqeqeshwa, futhi inguqulo entsha yemodeli ifunda “kulebula” ye iphutha kanye namarekhodi "agciniwe" asuka ekusakazweni.

Umfanekiso 7 Ukwenziwa kwerobhothi kwekhompuyutha ye-AI/ML ku-InterSystems IRIS

Uma kunesidingo sohlobo "lombono wesibili" ngekhwalithi yokutholwa okutholwe ngesikhathi kubalwa kwasendaweni ku-InterSystems IRIS, inqubo yomeluleki iyakhiwa ukuze kwenziwe ukuqeqeshwa nokusebenzisa amamodeli kudathasethi yokulawula kusetshenziswa izinsiza zamafu (isibonelo, iMicrosoft I-Azure, i-Amazon Web Services, i-Google Cloud Platform, njll.):

Umfanekiso 8 Umbono Wesibili ovela ku-Microsoft Azure ohlelwe yi-InterSystems IRIS

I-prototype yesimo sethu ku-InterSystems IRIS yakhelwe njengohlelo olususelwa kumenzeli lwezinqubo zokuhlaziya ezisebenzisana nento yemishini (ipompo), izindawo zokumodela zezibalo (Python, R kanye noJulia), futhi iqinisekise ukuzifundela kwakho konke okuhilelekile kwe-AI/ Izindlela ze-ML - ekusakazweni kwedatha yesikhathi sangempela .

Umfanekiso 9 Ukusebenza okuyinhloko kwesixazululo sesikhathi sangempela se-AI/ML ku-InterSystems IRIS

Umphumela osebenzayo we-prototype yethu:

• Isici sesampula esibonwa imodeli (Januwari 12):

• Isici esikhulayo esibonwa imodeli engazange ifakwe kusampula (Septhemba 11, iphutha ngokwalo likhonjwe ithimba labalungisayo ezinsukwini ezimbili kamuva, ngo-September 13):

Ukulingisa kudatha yangempela equkethe iziqephu ezimbalwa zesici esifanayo kubonise ukuthi isisombululo sethu, esisetshenziswe ku-platform ye-InterSystems IRIS, sisivumela ukuthi sibone ukuthuthukiswa kweziphambeko zalolu hlobo izinsuku ezimbalwa ngaphambi kokuba zitholwe ithimba lokulungisa.

I-InterSystems IRIS - inkundla yekhompyutha yesikhathi sangempela ye-AI/ML

Inkundla ye-InterSystems IRIS yenza kube lula ukuthuthukiswa, ukuthunyelwa kanye nokusebenza kwezixazululo zedatha yesikhathi sangempela. I-InterSystems IRIS iyakwazi ukwenza ngesikhathi esisodwa ukucutshungulwa kwedatha nokuhlaziya; sekela ukubukwa kwedatha evumelanisiwe ngokuvumelana namamodeli amaningi (okuhlanganisa ubuhlobo, isigaba, into kanye nedokhumenti); sebenza njengenkundla yokuhlanganisa imithombo eminingi yedatha nezinhlelo zokusebenza ngazinye; hlinzeka ngezibalo zesikhathi sangempela ezithuthukisiwe kudatha ehlelekile nengahlelekile. I-InterSystems IRIS iphinde inikeze izindlela zokusebenzisa amathuluzi okuhlaziya angaphandle futhi ivumela inhlanganisela evumelana nezimo yokubamba emafini naseziphakelini zendawo.

Izicelo ezakhelwe endaweni ye-InterSystems IRIS zisetshenziswa kuzo zonke izimboni ezihlukahlukene, zisiza izinkampani ukuthi zibone izinzuzo ezinkulu zezomnotho ngombono wamasu nokusebenza, ukwandisa ukuthathwa kwezinqumo ezinolwazi futhi kuvalwe igebe phakathi komcimbi, ukuhlaziya kanye nokwenza.

Umfanekiso we-10 InterSystems IRIS architecture kumongo we-AI/ML yesikhathi sangempela

Njengomdwebo odlule, umdwebo ongezansi uhlanganisa "uhlelo lokuxhumanisa" olusha (i-CD / CI / CT) nomdwebo wokugeleza kolwazi phakathi kwezinto zokusebenza zesikhulumi. Ukubuka kuqala nge-macromechanism CD futhi kuqhubeke nge-macromechanisms CI ne-CT.

Umfanekiso we-11 wolwazi ugeleza phakathi kwezinto ze-AI/ML zeplathifomu ye-InterSystems IRIS

Ingqikithi yomshini we-CD ku-InterSystems IRIS: abasebenzisi bepulatifomu (abathuthukisi bezixazululo ze-AI/ML) bazivumelanisa nezimo ezikhona kanye/noma badale ukuthuthukiswa okusha kwe-AI/ML besebenzisa umhleli wekhodi okhethekile wezindlela ze-AI/ML: Jupyter (igama eligcwele: Jupyter Notebook; ngokufushane, imibhalo edalwe kulo mhleli nayo ngezinye izikhathi ibizwa ngokuthi). Ku-Jupyter, umthuthukisi unethuba lokubhala, ukulungisa iphutha nokuqinisekisa ukusebenza (okuhlanganisa nokusebenzisa izithombe) kokuthuthukiswa okuthile kwe-AI/ML ngaphambi kokuthi kubekwe (“kusetshenziswe”) ku-InterSystems IRIS. Kuyacaca ukuthi ukuthuthukiswa okusha okudalwe ngale ndlela kuzothola ukulungiswa okuyisisekelo kuphela (njengoba, ikakhulukazi, i-Jupyter ayisebenzi nemifudlana yedatha yesikhathi sangempela) - lokhu kukuhleleka kwezinto, ngoba umphumela oyinhloko wentuthuko ku-Jupyter siwukuqinisekisa kokusebenza okubalulekile kwendlela ehlukile ye-AI/ML (“ibonisa umphumela olindelekile kusampula yedatha”). Ngokufanayo, indlela esivele ibekwe endaweni yesikhulumi (bona ama-macro-mechanisms alandelayo) ngaphambi kokulungisa iphutha ku-Jupyter ingase idinge “ukubuyisela emuva” kufomu “langaphambi kwenkundla” (ukufunda idatha kumafayela, ukusebenza ngedatha nge-xDBC esikhundleni samatafula, ukusebenzisana okuqondile nama-globals - ama-multidimensional data arrays InterSystems IRIS – njll.).

Isici esibalulekile sokuqaliswa kwe-CD ku-InterSystems IRIS: ukuhlanganiswa kwe-bidirectional kuye kwaqaliswa phakathi kweplatifomu ne-Jupyter, okuvumela okuqukethwe ku-Python, R no-Julia ukuthi kudluliselwe endaweni yesikhulumi (futhi, kamuva, kusetshenzwe endaweni yesikhulumi) (zonke ezintathu ziyizinhlelo izilimi ngezilimi zomthombo ovulekile ezihamba phambili ezihambisanayo). izindawo zokumodela zezibalo). Ngakho-ke, abathuthukisi bokuqukethwe be-AI/ML banethuba lokwenza “ukuthunyelwa okuqhubekayo” kwalokhu okuqukethwe endaweni yesikhulumi, besebenza kumhleli wabo we-Jupyter abawujwayele, onemitapo yolwazi ejwayelekile etholakala ePython, R, Julia, kanye nokwenza ukulungisa iphutha okuyisisekelo (uma kunesidingo) ngaphandle komsamo .

Masiqhubekele kumshini omkhulu we-CI ku-InterSystems IRIS. Umdwebo ubonisa inqubo enkulu ye-"robotizer yesikhathi sangempela" (inkimbinkimbi yezakhiwo zedatha, izinqubo zebhizinisi nezingcezu zekhodi ezihlelwe yibona ngezilimi zezibalo kanye ne-ObjectScript - ulimi lomdabu lokuthuthukiswa kwe-InterSystems IRIS). Umsebenzi wale nqubo enkulu ukugcina imigqa yedatha edingekayo ekusebenzeni kwezindlela ze-AI/ML (ngokusekelwe emifudlaneni yedatha edluliselwa endaweni yesikhulumi ngesikhathi sangempela), yenza izinqumo mayelana nokulandelana kwesicelo kanye "nokuhlukahluka" kwe-AI/ Izindlela ze-ML (ziphinde zibe “ama-algorithms ezibalo”, “amamodeli”, njll. - zingabizwa ngendlela ehlukile kuye ngokucaciswa kokusetshenziswa kanye nezintandokazi zetemu), gcina izinhlaka zedatha zisesikhathini samanje ukuze kuhlaziywe imiphumela yomsebenzi we-AI/ Izindlela ze-ML (amakhyubhu, amathebula, ama-multidimensional data array, njll.). njll. - okwemibiko, amadeshibhodi, njll.).

Isici esibalulekile sokuqaliswa kwe-CI ku-InterSystems IRIS: ukuhlanganiswa kwe-bidirectional kuye kwasetshenziswa phakathi kweplatifomu kanye nezindawo zokumodela zezibalo, okukuvumela ukuthi wenze okuqukethwe okusingathwe endaweni yesikhulumi ePython, R no-Julia ezindaweni zabo ngokulandelana futhi wamukele imiphumela yokubulawa. Lokhu kuhlanganiswa kwenziwa kokubili “ngemodi yesiphetho” (okungukuthi, okuqukethwe kwe-AI/ML kwakhiwa njengekhodi ye-ObjectScript eyenza izingcingo endaweni ezungezile) kanye “nakwimodi yenqubo yebhizinisi” (okungukuthi, okuqukethwe kwe-AI/ML kwakhiwa njengenqubo yebhizinisi usebenzisa umhleli wesithombe, noma ngezinye izikhathi usebenzisa i-Jupyter, noma usebenzisa i-IDE - IRIS Studio, Eclipse, Visual Studio Code). Ukutholakala kwezinqubo zebhizinisi zokuhlela ku-Jupyter kubonakala ngokuxhumana phakathi kwe-IRIS ezingeni le-CI kanye ne-Jupyter ezingeni le-CD. Ukubuka konke okunemininingwane eminingi yokuhlanganiswa nezimo zokumodela zezibalo kunikezwe ngezansi. Kulesi sigaba, ngokubona kwethu, kukhona zonke izizathu zokuqinisekisa ukuthi inkundla inawo wonke amathuluzi adingekayo okusebenzisa “ukuhlanganiswa okuqhubekayo” kwentuthuko ye-AI/ML (ephuma “ekuhanjisweni okuqhubekayo”) ibe yizixazululo zesikhathi sangempela ze-AI/ML.

Futhi indlela eyinhloko enkulu: CT. Ngaphandle kwayo, ngeke kube khona inkundla ye-AI/ML (yize “isikhathi sangempela” sizosetshenziswa nge-CD/CI). Ingqikithi ye-CT umsebenzi weplathifomu “onobuciko” bokufunda komshini kanye nobuhlakani bokwenziwa ngokuqondile ngezikhathi zokusebenza zezindawo zokumodela zezibalo: amamodeli, amatafula okusabalalisa, ama-matrix vectors, izendlalelo zamanethiwekhi e-neural, njll. Lo “msebenzi”, ezimweni eziningi, uqukethe ukudala izinto zobuciko ezishiwo ezindaweni (esimweni samamodeli, ngokwesibonelo, "indalo" iqukethe ukusetha ukucaciswa kwemodeli nokukhethwa okulandelayo kwamanani emingcele yayo - lokho okubizwa ngokuthi "ukuqeqeshwa" kwemodeli), ukusetshenziswa kwabo (kumamodeli: ukubala ngosizo lwabo "imodeli" yamanani okuguquguqukayo okuhlosiwe - izibikezelo, ubulungu besigaba, amathuba omcimbi, njll.) kanye nokuthuthukiswa kwalokho kakade ama-artifact adaliwe futhi asetshenziswa (isibonelo, ukuchaza kabusha isethi yokuhlukahluka kokufaka okuyimodeli ngokusekelwe emiphumeleni yohlelo lokusebenza - ukuze kuthuthukiswe ukunemba kokubikezela, njengenketho). Iphuzu elibalulekile ekuqondeni indima ye-CT "ukukhipha" kwayo ezintweni ezingokoqobo ze-CD ne-CI: I-CT izosebenzisa wonke ama-artifacts, igxile emininingwaneni yokubala neyezibalo yesisombululo se-AI/ML ngaphakathi kwamakhono ahlinzekwa yizindawo ezithile. Isibopho "sokunikeza okokufaka" kanye "nokulethwa kwemiphumela" kuzoba yisibopho se-CD ne-CI.

Isici esibalulekile sokuqaliswa kwe-CT ikakhulukazi ku-InterSystems IRIS: kusetshenziswa ukuhlanganiswa nezindawo zokumodela zezibalo esezishiwo ngenhla, inkundla inamandla okukhipha lawo ma-artifacts kumaseshini okusebenza asebenza ngaphansi kokulawula kwawo ezindaweni zezibalo kanye (okubaluleke kakhulu) ukujika. zibe izinto zedatha yesikhulumi. Isibonelo, ithebula lokusabalalisa elisanda kwakhiwa kuseshini yePython esebenzayo (ngaphandle kokumisa iseshini yePython) lidluliselwe endaweni yesikhulumi ngendlela, ngokwesibonelo, i-global (uhlu lwedatha ye-InterSystems IRIS) - futhi isetshenziswe. ngezibalo kwenye indlela ye-AI/ML- (esetshenziswa ngolimi lwenye indawo - isibonelo, ngo-R) - noma ithebula elibonakalayo. Esinye isibonelo: ngokuhambisana “nemodi evamile” yokusebenza kwemodeli (ngeseshini yokusebenza ye-Python), “i-auto-ML” yenziwa kudatha yayo yokufaka: ukukhethwa okuzenzakalelayo kokuguquguquka kokufakwayo okuphelele namanani epharamitha. Futhi kanye nokuqeqeshwa "okuvamile", imodeli ekhiqizayo ngesikhathi sangempela iphinde ithole "isiphakamiso sokwenza kahle" kokucaciswa kwayo - lapho isethi yokufaka okuguquguqukayo ishintsha, amanani wepharamitha ashintsha (akusekho ngenxa yokuqeqeshwa. kuPython, kodwa ngenxa yokuqeqeshwa ngenguqulo “ehlukile” ” yayo ngokwayo, njengesitaki se-H2O), okuvumela isisombululo se-AI/ML sisonke ukuthi sibhekane ngokuzimele nezinguquko ezingalindelekile kumvelo wedatha yokufaka kanye nezimo ezimodelayo. .

Ake sijwayelane kabanzi nokusebenza kweplatifomu ye-AI/ML ye-InterSystems IRIS, sisebenzisa isibonelo se-prototype yempilo yangempela.

Kumdwebo ongezansi, ohlangothini lwesobunxele lwesilayidi kunengxenye yenqubo yebhizinisi esebenzisa ukukhishwa kwemibhalo ku-Python ne-R. Engxenyeni emaphakathi kukhona amalogi abonakalayo okwenziwa kwezinye zalezi zikripthi, ngokulandelanayo, ku-Python kanye no-R. Ngokushesha ngemva kwazo kunezibonelo zokuqukethwe kolunye ulimi nolunye ulimi, okudluliselwa ukuze kusetshenziswe ezindaweni ezifanele. Ekugcineni kwesokudla kukhona okubukwayo okusekelwe emiphumeleni yokwenziwa kombhalo. Ukubukwa okuphezulu kwenziwa ku-IRIS Analytics (idatha ithathwe ku-Python yayiswa ku-InterSystems IRIS yesikhulumi sedatha futhi yaboniswa kudeshibhodi kusetshenziswa inkundla), ezansi zenziwa ngokuqondile kuseshini yomsebenzi ka-R futhi okukhiphayo kusuka lapho kuya kumafayela ayingcaca. . Isici esibalulekile: ucezu olwethuliwe ku-prototype lubophezelekile ekuqeqesheni imodeli (ukuhlelwa kwezimo zemishini) kudatha etholwe ngesikhathi sangempela kusuka kunqubo yesifanisi semishini, phezu komyalo ovela enqubweni yokuqapha ikhwalithi yokuhlukanisa ebonwe ngesikhathi sokusetshenziswa kwemodeli. Ukuqaliswa kwesisombululo se-AI/ML ngendlela yesethi yezinqubo zokusebenzisana (“ama-ejenti”) kuzoxoxwa ngakho kabanzi.

Umfanekiso we-12 Ukusebenzisana nePython, R noJulia ku-InterSystems IRIS

Izinqubo zeplathifomu (ziphinde “ziyizinqubo zebhizinisi”, “izinqubo zokuhlaziya”, “amapayipi”, njll. - kuye komongo), okokuqala, zihlelwa kumhleli wenqubo yebhizinisi eliyisithombe endaweni ngokwayo, futhi kuleso simo. indlela kokubili umdwebo webhulokhi kanye nendlela ye-AI/ML ehambisanayo (ikhodi yohlelo) edalwa kanyekanye. Uma sithi "indlela ye-AI/ML itholakele," ekuqaleni sisho ukuxutshwa (ngaphakathi kwenqubo eyodwa): okuqukethwe ezilimini zezindawo zokumodela zezibalo kuncikene nokuqukethwe ku-SQL (okuhlanganisa nezandiso ezivela I-IntegratedML), ku-InterSystems ObjectScript, nezinye izilimi ezisekelwayo. Ngaphezu kwalokho, inqubo yesikhulumi ihlinzeka ngamathuba abanzi kakhulu "okunikezela" ngendlela yezingcezu ezibekwe ngokohlelo (njengoba kungabonwa esibonelweni esikumdwebo ongezansi), okuvumela ukuthi uhlele ngempumelelo ngisho nokuqukethwe okuyinkimbinkimbi kakhulu ngaphandle kokuthi "uqhamuke" yefomethi yesithombe (kumafomethi “okungewona wesithombe”). »izindlela/amakilasi/izinqubo, njll.). Okusho ukuthi, uma kunesidingo (futhi kubonakala kumaphrojekthi amaningi), sonke isixazululo se-AI/ML singasetshenziswa ngefomethi yokuzibhala eyisithombe. Sicela uqaphele ukuthi engxenyeni emaphakathi yomdwebo ongezansi, omelela "izinga eliphezulu lokuzalela", kusobala ukuthi ngaphezu komsebenzi wangempela wokuqeqesha imodeli (usebenzisa i-Python ne-R), ukuhlaziywa kwalokho okubizwa ngokuthi. Ijika le-ROC lemodeli eqeqeshiwe liyanezelwa, livumela ngokubonakalayo (futhi ngekhompuyutha futhi) ukuhlola ikhwalithi yokuqeqeshwa - futhi lokhu kuhlaziya kwenziwa ngolimi lukaJulia (olwenziwa, ngokufanelekile, endaweni yezibalo kaJulia).

Umfanekiso we-13 Indawo ebonakalayo yokwakhiwa kwezixazululo ze-AI/ML ku-InterSystems IRIS

Njengoba kushiwo ngaphambili, ukuthuthukiswa kokuqala kanye (kwezinye izimo) ukulungiswa kwezinqubo ze-AI/ML esezivele zisetshenziswe endaweni yesikhulumi kuzo/kungenziwa ngaphandle kweplatifomu kumhleli we-Jupyter. Emdwebeni ongezansi sibona isibonelo sokushintsha inqubo yesikhulumi esikhona (efana nasomdwebo ongenhla) - yile ndlela ucezu olunomthwalo wemfanelo wokuqeqesha imodeli ku-Jupyter. Okuqukethwe kwe-Python kuyatholakala ukuze kuhlelwe, kulungiswe iphutha, futhi kukhishwe ihluzo ngqo ku-Jupyter. Izinguquko (uma kudingekile) zingenziwa ngokuvumelanisa okusheshayo kunqubo yesikhulumi, okuhlanganisa inguqulo yayo ekhiqizayo. Okuqukethwe okusha kungadluliswa endaweni yesikhulumi ngendlela efanayo (inqubo entsha yesikhulumi ikhiqizwa ngokuzenzakalelayo).

Umfanekiso we-14 Ukusebenzisa i-Jupyter Notebook ukuhlela injini ye-AI/ML ku-InterSystems IRIS yesikhulumi

Ukuguqulwa kwenqubo yesikhulumi kungenziwa hhayi kuphela ngefomethi yesithombe noma ikhompuyutha ephathekayo - kodwa nangefomethi "yonke" ye-IDE (Indawo Yokuthuthukiswa Okudidiyelwe). Lawa ma-IDE yi-IRIS Studio (isitudiyo se-IRIS somdabu), Ikhodi ye-Visual Studio (isandiso se-InterSystems IRIS se-VSCode) kanye ne-Eclipse (i-plugin ye-Atelier). Kwezinye izimo, kuyenzeka ukuthi ithimba labathuthukisi lisebenzise wonke ama-IDE amathathu ngesikhathi esisodwa. Umdwebo ongezansi ubonisa isibonelo sokuhlela inqubo efanayo ku-IRIS studio, ku-Visual Studio Code kanye naku-Eclipse. Impela konke okuqukethwe kuyatholakala ukuze kuhlelwe: Python/R/Julia/SQL, ObjectScript, kanye nenqubo yebhizinisi.

Umfanekiso we-15 Ukuthuthukiswa kwenqubo yebhizinisi ye-InterSystems IRIS kuma-IDE ahlukahlukene

Amathuluzi okuchaza nokusebenzisa izinqubo zebhizinisi ze-InterSystems IRIS Olimini Lwenqubo Yebhizinisi (BPL) afanelwe ukushiwo ngokukhethekile. I-BPL yenza kube nokwenzeka ukusebenzisa "izingxenye zokuhlanganisa esezivele zenziwe" (imisebenzi) ezinqubweni zebhizinisi - empeleni, ezinikeza zonke izizathu zokusho ukuthi "ukuhlanganiswa okuqhubekayo" kuqaliswa ku-InterSystems IRIS. Izingxenye zenqubo yebhizinisi esezilungile (imisebenzi nokuxhumana phakathi kwazo) ziyi-accelerator enamandla yokuhlanganisa isisombululo se-AI/ML. Futhi hhayi imihlangano kuphela: ngenxa yemisebenzi nokuxhumana phakathi kwayo phezu kwentuthuko ehlukene ye-AI/ML nezindlela, kuvela “ungqimba oluzimele lokuphatha”, olukwazi ukwenza izinqumo ngokuya ngesimo, ngesikhathi sangempela.

Umfanekiso we-16 Izingxenye zenqubo yebhizinisi esezilungele ukuhlanganiswa okuqhubekayo (CI) ku-InterSystems IRIS yesikhulumi

Umqondo wezinhlelo zama-ejenti (owaziwa nangokuthi "izinhlelo zama-ejenti amaningi") unesikhundla esiqinile ekwenziweni kwamarobhothi, futhi inkundla ye-InterSystems IRIS isekela ngokuphilayo ngokwakhiwa "kwenqubo yomkhiqizo". Ngaphezu kwamathuba angenamkhawulo "okufaka" inqubo ngayinye ngokusebenza okudingekayo kusixazululo sisonke, ukunikeza uhlelo lwezinqubo zeplatifomu ngempahla "ye-ejensi" kukuvumela ukuthi udale izixazululo ezisebenzayo zezenzakalo ezilingisayo ezingazinzile kakhulu (ukuziphatha komphakathi/ amasistimu ezinto eziphilayo, izinqubo zobuchwepheshe ezibonakala kancane, njll.).

Umfanekiso 16 Ukusebenza kwesisombululo se-AI/ML njengohlelo lwenqubo yebhizinisi olususelwe kumenzeli ku-InterSystems IRIS

Siqhubeka nokubuyekezwa kwethu kwe-InterSystems IRIS ngendaba emayelana nokusetshenziswa kwenkundla yokuxazulula wonke amakilasi ezinkinga zesikhathi sangempela (isingeniso esinemininingwane efanelekile kwezinye zezindlela ezinhle kakhulu zeplathifomu AI/ML ku-InterSystems IRIS ingatholakala endaweni eyodwa. okwedlule kwethu ama-webinars).

Okushisayo ezithendeni zomdwebo odlule, ngezansi umdwebo onemininingwane eminingi yesistimu ye-ejenti. Umdwebo ubonisa i-prototype efanayo, zonke izinqubo ezine zama-ejenti ziyabonakala, ubudlelwano phakathi kwazo zidwetshwe ngohlelo: I-GENERATOR - icubungula ukudalwa kwedatha ngezinzwa zemishini, BUFFER - ilawula olayini bedatha, I-ANALYZER - yenza ukufunda komshini ngokwayo, I-MONITOR - iqapha ikhwalithi yokufunda komshini futhi ihambisa isignali mayelana nesidingo sokuqeqesha kabusha imodeli.

Umfanekiso 17 Ukwakhiwa kwesisombululo se-AI/ML ngendlela yohlelo lwenqubo yebhizinisi esekwe kumenzeli ku-InterSystems IRIS

Umdwebo ongezansi ubonisa ukusebenza okuzimele kwesinye isibonelo serobhothi (ukubonwa kombala wemizwa wemibhalo) isikhathi esithile. Engxenyeni engenhla ukuvela kwenkomba yekhwalithi yokuqeqeshwa kwemodeli (ikhwalithi ikhula), engxenyeni engezansi yi-dynamics yenkomba yekhwalithi yesicelo semodeli kanye namaqiniso okuqeqeshwa okuphindaphindiwe (imivimbo ebomvu). Njengoba ubona, isixazululo sizifundele ngokuphumelelayo nangokuzimele, futhi sisebenza ezingeni elinikeziwe lekhwalithi (amanani wekhwalithi wekhwalithi awehli ngaphansi kwama-80%).

Umfanekiso we-18 Ukuqeqeshwa okuqhubekayo (self-) (CT) ku-InterSystems IRIS yesikhulumi

Siphinde sasho “i-auto-ML” ngaphambili, kodwa umdwebo ongezansi ubonisa ukusetshenziswa kwalokhu kusebenza ngokuningiliziwe kusetshenziswa isibonelo somunye umfuziselo. Umdwebo wesithombe socezu lwenqubo yebhizinisi ubonisa umsebenzi oqalisa ukumodela kusitaki se-H2O, ubonisa imiphumela yalokhu kumodela (ukubusa okucacile kwemodeli ewumphumela phezu kwamamodeli “enziwe umuntu,” ngokuya ngomdwebo wokuqhathanisa we Amajika e-ROC, kanye nokuhlonza okuzenzakalelayo “kokuguquguquka okunamandla kakhulu” okutholakala kusethi yedatha yoqobo). Iphuzu elibalulekile lapha ukonga isikhathi nezinsiza ezingochwepheshe okufinyelelwa “nge-auto-ML”: lokho okwenziwa inqubo yethu yesikhulumi ngesigamu somzuzu (ukuthola nokuqeqesha imodeli efanelekile) kungathatha uchwepheshe kusukela ngesonto kuya enyangeni.

Umfanekiso we-19 Ukuhlanganiswa kwe- "auto-ML" kusisombululo se-AI/ML ku-InterSystems IRIS yesikhulumi

Umdwebo ongezansi ulahlekelwa iphuzu kancane, kodwa kuyindlela enhle yokuqeda indaba mayelana nezigaba zezinkinga zesikhathi sangempela ezixazululwa: siyakukhumbuza ukuthi ngawo wonke amakhono we-InterSystems IRIS yesikhulumi, amamodeli okuqeqesha ngaphansi kokulawula kwawo akuphoqelekile. Inkundla ingathola ngaphandle lokho okubizwa nge-PMML okucaciswayo kwemodeli, eqeqeshelwe ithuluzi elingekho ngaphansi kokulawulwa kwenkundla - futhi isebenzise le modeli ngesikhathi sangempela kusukela ngesikhathi ithengwa ngayo. Imininingwane ye-PMML. Kubalulekile ukunaka ukuthi akuwona wonke ama-artifact we-AI/ML angancishiswa abe esimisweni se-PMML, noma ngabe iningi lama-artifact ajwayelekile likuvumela lokhu. Ngakho, isiteji se-InterSystems IRIS "i-loop evulekile" futhi akusho "ubugqila bepulatifomu" kubasebenzisi.

Umfanekiso we-20 Ukuhlanganiswa kwe- "auto-ML" kusisombululo se-AI/ML ku-InterSystems IRIS yesikhulumi

Ake siklelise izinzuzo zeplathifomu ezengeziwe ze-InterSystems IRIS (ukuze kucace, maqondana nokulawulwa kwenqubo), ezibaluleke kakhulu ekuzizenzeleni kobuhlakani bokwenziwa nokufunda komshini kwesikhathi sangempela:

• Kuthuthukiswe amathuluzi okuhlanganiswa nanoma yimiphi imithombo yedatha nabathengi (uhlelo lokulawula inqubo/i-SCADA, okokusebenza, i-MRO, i-ERP, njll.)
• Kwakhelwe ngaphakathi I-DBMS yamamodeli amaningi ukusebenza okuphezulu kokwenziwayo nokucubungula (i-Hybrid Transaction/Analytical Processing, HTAP) yanoma iyiphi ivolumu yedatha yenqubo yezobuchwepheshe
• Amathuluzi okuthuthukisa ukuthunyelwa okuqhubekayo kwezinjini ze-AI/ML zezixazululo zesikhathi sangempela ezisekelwe kuPython, R, Julia
• Izinqubo zebhizinisi eziguquguqukayo zokuhlanganiswa okuqhubekayo kanye (nokwakho) ukufunda izinjini zesixazululo ze-AI/ML zesikhathi sangempela
• Amathuluzi Akhelwe ngaphakathi EBusiness Intelligence okubona ngeso lengqondo idatha yenqubo kanye nemiphumela yesisombululo se-AI/ML
• Ukuphathwa kwe-API ngokuletha imiphumela yesisombululo se-AI/ML ukuze kucutshungulwe amasistimu okulawula/i-SCADA, ulwazi nezinhlelo zokuhlaziya, ukuthumela izexwayiso, njll.

Izixazululo ze-AI/ML kuplathifomu ye-InterSystems IRIS zingena kalula kungqalasizinda ye-IT ekhona. Inkundla ye-InterSystems IRIS iqinisekisa ukwethembeka okuphezulu kwezixazululo ze-AI/ML ngokusekela ukumiswa okubekezelela amaphutha nokubekezelela izinhlekelele kanye nokuthunyelwa okuguquguqukayo ezindaweni ezibonakalayo, kumaseva abonakalayo, emafwini angasese nawomphakathi, kanye neziqukathi ze-Docker.

Ngakho-ke, i-InterSystems IRIS iyinkundla yekhompyutha ye-AI/ML yesikhathi sangempela yendawo yonke. Ukuma kwendawo yonke yesikhulumi sethu kuqinisekiswa ngokusebenza ngokungabi bikho kwemikhawulo ye-de facto ebunzimeni bezibalo ezisetshenzisiwe, ikhono le-InterSystems IRIS ukuhlanganisa (ngesikhathi sangempela) ukucutshungulwa kwezimo ezivela ezinhlobonhlobo zezimboni, kanye nokuzivumelanisa nezimo okuyingqayizivele noma yimiphi imisebenzi yesikhulumi nezindlela ezihambisana nezidingo ezithile zabasebenzisi.

Umfanekiso 21 InterSystems IRIS - inkundla yekhompyutha yesikhathi sangempela ye-AI/ML

Ukuze uthole ukusebenzelana okunamandla nalabo babafundi bethu abathanda ukwaziswa okwethulwe lapha, sincoma ukuthi ungazigcini ngokukufunda futhi uqhubeke nengxoxo “bukhoma.” Sizokujabulela ukukunikeza ukwesekwa ngokwakhiwa kwezimo zesikhathi sangempela ze-AI/ML ngokuphathelene nemininingwane yenkampani yakho, senze i-prototyping ehlanganyelwe endaweni yesikhulumi se-InterSystems IRIS, senze futhi sisebenzise ngokuzijwayeza umhlahlandlela wokwethula ubuhlakani bokwenziwa nokufunda ngomshini. ezinqubweni zakho zokukhiqiza nezokuphatha. I-imeyili Yethu Yokuxhumana Yethimba Lochwepheshe be-AI/ML - [i-imeyili ivikelwe].

Source: www.habr.com