InterSystems IRIS - dandamali na AI / ML na ainihi na duniya

Mawallafi: Sergey Lukyanchikov, injiniya mai ba da shawara a InterSystems

Kiran lissafin AI/ML na ainihin lokaci

Bari mu fara da misalai daga ƙwarewar aikin Kimiyyar Bayanai a InterSystems:

• An haɗa tashar tashar mai siye da aka ɗora zuwa tsarin shawarwarin kan layi. Za a yi gyare-gyaren tallace-tallace a cikin hanyar sadarwar tallace-tallace (misali, maimakon layin tallace-tallace na "lebur", yanzu za a yi amfani da matrix "ɓangarorin-dabarun"). Menene ya faru da injunan shawarwari? Menene zai faru da ƙaddamarwa da sabunta bayanai ga injin shawarwarin (ƙarar bayanan shigarwa ya karu da sau 25000)? Menene ya faru da ci gaban shawarwarin (buƙatar rage madaidaicin tacewa na sharuɗɗan shawarwarin da dubunnan dubunnan saboda karuwar adadin su da "kewayon")?
• Akwai tsarin kula da yiwuwar lahani da ke tasowa a cikin kayan aikin. An haɗa tsarin sarrafa tsari mai sarrafa kansa zuwa tsarin sa ido, yana watsa dubban sigogin tsarin fasaha kowane daƙiƙa. Menene zai faru da tsarin kulawa wanda a baya yayi aiki akan "samfurori na hannu" (yana iya samar da sa ido na yiwuwar na biyu zuwa na biyu)? Abin da zai faru idan sabon toshe na ɗaruruwan ginshiƙai ya bayyana a cikin bayanan shigarwa tare da karantawa daga na'urori masu auna firikwensin da aka ƙara kwanan nan zuwa tsarin sarrafa tsari (zai zama dole kuma tsawon lokacin da za a dakatar da tsarin kulawa don haɗa bayanai daga sabbin na'urori masu auna firikwensin a cikin bincike). )?
• An ƙirƙiri saitin hanyoyin AI / ML (shawarwari, saka idanu, tsinkaya) waɗanda ke amfani da sakamakon aikin juna. Awanni nawa ne ake buƙata kowane wata don daidaita aikin wannan rukunin zuwa canje-canjen bayanan shigarwa? Menene "jinkirin" gabaɗaya lokacin da goyan bayan hadaddun yanke shawara na gudanarwa (yawan faruwar sabbin bayanan tallafi a cikinsa dangane da yawan faruwar sabbin bayanan shigarwa)?

Taƙaice waɗannan da wasu misalai da yawa, mun zo ga ƙirƙira ƙalubalen da ke tasowa lokacin da aka matsa zuwa yin amfani da na'ura da kuma hanyoyin fasaha na wucin gadi a ainihin lokacin:

• Shin mun gamsu da saurin halitta da daidaitawa (zuwa yanayin canzawa) na ci gaban AI / ML a cikin kamfaninmu?
• Nawa ne hanyoyin AI/ML da muke amfani da su ke tallafawa gudanar da kasuwanci na lokaci-lokaci?
• Shin hanyoyin AI / ML da muke amfani da su suna iya dacewa da kansu (ba tare da masu haɓakawa ba) su dace da canje-canje a cikin bayanai da ayyukan sarrafa kasuwanci?

Labarinmu shine cikakken bayyani na iyawar dandali na InterSystems IRIS dangane da tallafin duniya don ƙaddamar da hanyoyin AI / ML, taro (haɗin kai) hanyoyin AI / ML, da horo (gwaji) na AI / ML mafita a kan m. rafukan bayanai. Za mu dubi binciken kasuwa, nazarin shari'ar AI/ML mafita, da kuma ra'ayi na abin da muke kira dandali na AI / ML na ainihi a cikin wannan labarin.

Abin da muka sani daga safiyo: aikace-aikace na lokaci-lokaci

Результаты bincikenwanda aka gudanar tsakanin kusan kwararrun IT 800 a cikin 2019 ta Lightbend, suna magana da kansu:

Hoto 1 Jagoran masu amfani da bayanan ainihin lokacin

Bari mu kawo wasu mahimman gutsuttsura rahoton akan sakamakon wannan binciken a cikin fassarar mu:

"... Abubuwan da ke faruwa a cikin shahararrun kayan aiki don haɗawa da rafukan bayanai da kuma, a lokaci guda, tallafawa ƙididdiga a cikin kwantena suna ba da amsawar haɗin gwiwa ga buƙatar kasuwa don ƙarin amsawa, ma'ana, tsari mai mahimmanci na mafita mai mahimmanci. Bayanan yawo yana canja bayanai cikin sauri fiye da bayanan fakiti na gargajiya. Ƙara zuwa wannan shine ikon yin amfani da sauri da hanyoyin ƙididdiga, kamar, alal misali, shawarwarin tushen AI / ML, ƙirƙirar fa'idodi masu fa'ida ta hanyar haɓaka gamsuwar abokin ciniki. Hakanan tseren ƙarfin hali yana tasiri duk wani matsayi a cikin tsarin DevOps - samar da haɓaka aikace-aikace da turawa mafi inganci. … ƙwararrun IT ɗari takwas da huɗu sun ba da bayanai game da amfani da bayanan da ke gudana a cikin ƙungiyoyin su. Masu amsa sun fi yawa a cikin ƙasashen Yamma (41% a Turai da 37% a Arewacin Amirka) kuma kusan an rarraba su tsakanin ƙanana, matsakaita da manyan kamfanoni. ...

... Hankali na wucin gadi ba talla ba ne. Kashi XNUMX cikin XNUMX na waɗanda suka riga sun yi amfani da sarrafa bayanan rafi a cikin aikace-aikacen AI / ML masu aiki sun tabbatar da cewa amfani da AI / ML zai ga mafi girma girma a cikin shekara mai zuwa (idan aka kwatanta da sauran aikace-aikace).

• Bisa ga yawancin masu amsawa, yin amfani da bayanan bayanai a cikin yanayin AI / ML zai ga mafi girma girma a cikin shekara mai zuwa.
• Aikace-aikace a cikin AI/ML za su yi girma ba kawai saboda ingantattun nau'ikan al'amura ba, har ma saboda al'amuran gargajiya waɗanda ake ƙara amfani da bayanan lokaci na ainihi.
• Baya ga AI / ML, matakin sha'awar masu amfani da bututun bayanai na IoT yana da ban sha'awa - 48% na waɗanda suka riga sun haɗa bayanan IoT sun ce aiwatar da yanayin kan wannan bayanan zai ga ƙaruwa sosai nan gaba. ..."

Daga wannan binciken mai ban sha'awa mai ban sha'awa, a bayyane yake cewa hangen nesa na koyon injin da yanayin hankali na wucin gadi a matsayin jagorori a cikin amfani da rafukan bayanai sun riga sun "kan hanya." Amma mahimmin abin lura daidai shine fahimtar AI/ML na ainihi ta hanyar ruwan tabarau na DevOps: anan zamu iya fara magana game da sauyin al'adun da har yanzu ke da rinjaye na "AI / ML da za a iya zubarwa tare da cikakken saitin bayanai."

Manufar dandamali na AI/ML na ainihi

Wani yanki na aikace-aikace na ainihin AI/ML shine sarrafa tsari a masana'antu. Yin amfani da misalinta da kuma yin la'akari da tunanin da suka gabata, za mu tsara manufar dandamali na AI/ML na ainihi.
Amfani da hankali na wucin gadi da koyan injina a cikin sarrafa tsari yana da fasali da yawa:

• Ana karɓar bayanai game da yanayin tsarin fasaha da ƙarfi: tare da mitar mai girma kuma don nau'ikan sigogi masu yawa (har zuwa dubun dubatar siga waɗanda aka watsa ta sakan ɗaya daga tsarin sarrafa tsari)
• Bayanai game da gano lahani, ba tare da ambaton bayanai game da ci gaban su ba, akasin haka, ba su da yawa kuma ba bisa ka'ida ba, suna nuna rashin isa ga lahani da kuma yanayin su a cikin lokaci (sau da yawa ana wakilta ta takardun takarda)
• Daga ra'ayi mai amfani, kawai "taga mai dacewa" na bayanan tushen yana samuwa don horarwa da amfani da samfuri, yana nuna yanayin tsarin fasaha akan madaidaicin tazarar zamewa yana ƙarewa tare da ƙimar karatun ƙarshe na sigogin tsari.

Waɗannan fasalulluka sun tilasta mana, ban da karɓa da aiki na asali a cikin ainihin lokacin "siginar shigarwar watsa shirye-shiryen watsa shirye-shirye" mai ƙarfi daga tsarin fasaha, yin (a layi ɗaya) aikace-aikacen, horo da sarrafa ingancin sakamakon AI / Samfuran ML - kuma a ainihin lokacin. "Frame" wanda samfuranmu "gani" a cikin taga mai zamewa na dacewa yana canzawa koyaushe - kuma tare da shi, ingancin sakamakon aikin AI / ML samfuran da aka horar akan ɗayan "firam" a baya kuma yana canzawa. . Idan ingancin sakamakon aikin AI / ML model ya tabarbare (alal misali: ƙimar rarrabuwar "alarm-alarm" ta wuce iyakokin da muka ayyana), ƙarin horo na samfuran ya kamata a ƙaddamar da shi ta atomatik akan. mafi halin yanzu "frame" - da kuma zabi na lokacin da za a kaddamar da ƙarin horo na model ya kamata la'akari da yadda da duration na horo da kanta, da kuzarin kawo cikas na deterioration a cikin ingancin aiki na halin yanzu version na model (tun da ana ci gaba da amfani da nau'ikan samfuran na yanzu yayin da ake horar da samfuran, kuma har sai an samar da nau'ikan ''sabbin horarwa'').

InterSystems IRIS yana da mabuɗin damar dandamali don ba da damar AI / ML mafita don sarrafa tsari na lokaci-lokaci. Ana iya raba waɗannan damar zuwa manyan ƙungiyoyi uku:

• Ci gaba da turawa (Ci gaba da Aiwatar / Bayarwa, CD) na sababbi ko daidaita hanyoyin AI / ML da ke akwai a cikin ingantaccen bayani mai aiki a ainihin lokacin akan dandamalin InterSystems IRIS
• Ci gaba da Haɗuwa (CI) a cikin mafita mai inganci guda ɗaya na kwararar bayanan tsarin fasaha mai shigowa, layin bayanai don aikace-aikacen / horo / kula da ingantattun hanyoyin AI / ML da musayar bayanai / lambar / ayyukan sarrafawa tare da yanayin ƙirar ƙira, waɗanda aka tsara a cikin ainihin lokaci. dandamali InterSystems IRIS
• Ci gaba (ci gaba da kai) horo (Ci gaba da horo, CT) na hanyoyin AI / ML, wanda aka yi a cikin yanayin ƙirar ƙididdiga ta amfani da bayanai, lamba da ayyukan sarrafawa ("yanke shawarar da aka yanke") wanda InterSystems IRIS dandamali ya watsa.

Rarraba damar dandali dangane da koyan na'ura da basirar wucin gadi cikin ainihin waɗannan ƙungiyoyi ba na haɗari ba ne. Bari mu faɗi hanyoyin bazawa Google, wanda ke ba da tushen ra'ayi don wannan rarrabuwa, a cikin fassararmu:

"... Ma'anar DevOps, sanannen kwanakin nan, ya shafi ci gaba da aiki na manyan tsarin bayanai. Abubuwan da ake amfani da su na aiwatar da wannan ra'ayi shine raguwa a cikin tsawon lokacin ci gaba na ci gaba, ƙaddamar da ci gaba da sauri, da sassauci a cikin shirin saki. Don cimma waɗannan fa'idodin, DevOps ya ƙunshi aiwatar da aƙalla ayyuka biyu:

• Cigaba da Haɗuwa (CI)
• Isar da Ci gaba (CD)

Hakanan waɗannan ayyukan suna amfani da dandamali na AI / ML don tabbatar da abin dogaro kuma mai aiwatarwa na mafita AI/ML mai albarka.

Dandalin AI/ML sun bambanta da sauran tsarin bayanai ta fuskoki masu zuwa:

• Ƙwararrun Ƙungiya: Lokacin ƙirƙirar mafita na AI / ML, ƙungiyar yawanci sun haɗa da masana kimiyyar bayanai ko ƙwararrun "ilimi" a fagen bincike na bayanai waɗanda ke gudanar da nazarin bayanai, haɓakawa da gwada samfuran. Waɗannan membobin ƙungiyar ƙila ba ƙwararrun ƙwararrun masu haɓaka lamba bane.
• Ci gaba: Injin AI/ML gwaji ne a cikin yanayi. Don magance matsala ta hanya mafi inganci, ya zama dole a bi ta hanyar haɗakarwa daban-daban na masu canjin shigarwa, algorithms, hanyoyin ƙirar ƙira da sigogin ƙira. Matsalolin irin wannan binciken yana cikin gano "abin da ya yi aiki / bai yi aiki ba", yana tabbatar da sake fasalin abubuwan da ke faruwa, haɓakar ci gaba don maimaita aiwatarwa.
• Gwaji: Gwajin AI/ML injuna na buƙatar gwaji mai faɗi fiye da sauran ci gaba. Baya ga daidaitattun naúrar da gwaje-gwajen haɗin kai, ana gwada ingancin bayanai da ingancin sakamakon amfani da samfurin zuwa horo da samfuran sarrafawa.
• Ƙaddamarwa: Ƙaddamar da AI/ML mafita bai iyakance ga ayyukan tsinkaya da ke amfani da samfurin da aka horar da sau ɗaya ba. AI / ML mafita an gina su a kusa da bututun matakai masu yawa waɗanda ke yin horon ƙirar ƙira da aikace-aikace. Aiwatar da irin waɗannan bututun ya haɗa da sarrafa ayyuka marasa kan gado bisa ga al'ada da masana kimiyyar bayanai suka yi da hannu don samun damar horarwa da gwada ƙira.
• Yawan aiki: Injin AI / ML na iya rasa yawan aiki ba kawai saboda ƙarancin shirye-shirye ba, har ma saboda canjin yanayin shigar da bayanai akai-akai. A wasu kalmomi, aikin hanyoyin AI / ML na iya raguwa saboda dalilai masu yawa fiye da ayyukan ci gaba na al'ada. Wanne yana haifar da buƙatar saka idanu (kan layi) aikin injunan AI / ML ɗin mu, da kuma aika faɗakarwa ko ƙin sakamako idan alamun aikin ba su cika tsammanin ba.

Hanyoyin AI / ML sun yi kama da sauran tsarin bayanai a cikin cewa duka biyu suna buƙatar ci gaba da haɗa lambar tare da sarrafa sigar, gwajin naúrar, gwajin haɗin kai, da ci gaba da ƙaddamar da ci gaba. Koyaya, game da AI/ML, akwai bambance-bambance masu mahimmanci da yawa:

• CI (Ci gaba da Haɗin kai) baya iyakance ga gwaji da tabbatar da lambar abubuwan da aka tura - ya kuma haɗa da gwaji da ingantattun bayanai da samfuran AI/ML.
• CD (Ci gaba da Bayarwa / Aiwatarwa, ci gaba da turawa) ba'a iyakance ga rubuce-rubuce da sakin fakiti ko ayyuka ba, amma yana nuna dandamali don haɗawa, horo da aikace-aikacen mafita na AI / ML.
• CT (Ci gaba da horo, ci gaba da horo) sabon abu ne [kimanin. marubucin labarin: wani sabon abu dangane da ra'ayin gargajiya na DevOps, wanda CT yake, a matsayin mai mulkin, Ci gaba da Gwaji], na asali a cikin AI / ML dandamali, alhakin gudanar da sarrafa kansa na hanyoyin horo da kuma amfani da AI. / ML model. ..."

Zamu iya bayyana cewa koyan na'ura da basirar wucin gadi da ke aiki akan bayanan ainihin-lokaci suna buƙatar ɗimbin kayan aiki da ƙwarewa (daga haɓaka lambar zuwa tsara yanayin ƙirar ƙira), kusanci kusa tsakanin duk wuraren aiki da batutuwa, ingantaccen tsarin ɗan adam da ingantaccen tsari. albarkatun inji.

Halin lokaci na ainihi: gane haɓakar lahani a cikin famfunan ciyarwa

Ci gaba da yin amfani da yankin kula da tsari a matsayin misali, yi la'akari da takamaiman matsala (mun riga an ambata a farkon): muna buƙatar samar da kulawa ta ainihi na ci gaban lahani a cikin famfo bisa la'akari da ma'auni na ƙimar tsari. da rahotanni daga ma'aikatan gyara game da lahani da aka gano.

Hoto 2 Tsarin matsala don lura da ci gaban lahani

Wani fasali na mafi yawan ayyukan da aka gabatar ta wannan hanyar a aikace shine cewa dole ne a yi la'akari da daidaito da ingancin karɓar bayanai (APCS) daidai da yanayin abubuwan da suka faru na al'ada da rashin daidaituwa (da rajista) na lahani iri-iri. A wasu kalmomi: bayanai daga tsarin sarrafa tsari suna zuwa sau ɗaya a cikin daƙiƙa guda, daidai kuma daidai, kuma ana yin bayanin kula game da lahani tare da fensir sinadari wanda ke nuna kwanan wata a cikin babban littafin rubutu a cikin taron bita (misali: “12.01 - leak cikin murfin. daga gefe na 3rd bearing).

Don haka, zamu iya ƙara ƙirar matsalar tare da ƙayyadaddun ƙayyadaddun mahimmanci masu zuwa: muna da "lakabin" ɗaya kawai na lahani na wani nau'i na musamman (watau misali na lahani na wani nau'i na musamman yana wakilta ta hanyar bayanai daga sarrafa tsari). tsarin akan takamaiman kwanan wata - kuma ba mu da ƙarin misalai na lahani na wannan nau'in musamman). Wannan ƙayyadaddun yana ɗaukar mu nan da nan fiye da ikon koyon injin na gargajiya (ilimin kulawa), wanda yakamata a sami “tags” da yawa.

Hoto 3 Bayyana aikin sa ido kan ci gaban lahani

Shin za mu iya "yawan" kawai "tag" a hannunmu? Eh za mu iya. Halin halin yanzu na famfo yana da alamar kamanni da lahani masu rijista. Ko da ba tare da amfani da hanyoyin ƙididdigewa ba, a matakin tsinkaye na gani, ta hanyar lura da haɓakar ƙimar bayanan da ke zuwa daga tsarin sarrafa tsari, zaku iya koya da yawa:

Hoto 4 Dynamics na yanayin famfo akan bangon "alama" na lahani na nau'in da aka bayar

Amma hangen nesa (aƙalla a yanzu) ba shine mafi dacewa janareta na “tags” a cikin yanayin mu mai saurin canzawa ba. Za mu kimanta kamancen yanayin famfo na yanzu zuwa lahani da aka ruwaito ta amfani da gwajin ƙididdiga.

Hoto na 5 Aiwatar da gwajin ƙididdiga zuwa bayanan da ke shigowa akan bangon “lakabin” lahani.

Gwajin ƙididdiga yana ƙayyade yiwuwar cewa rikodin tare da ƙimar ma'aunin tsarin fasaha a cikin "fakitin kwarara" da aka karɓa daga tsarin sarrafa tsari suna kama da bayanan "tag" na wani lahani na wani nau'i. Ƙimar yiwuwar (ƙididdigar kamanni na ƙididdiga) da aka ƙididdige sakamakon yin amfani da gwajin ƙididdiga an canza shi zuwa ƙimar 0 ko 1, zama "lakabi" don koyon na'ura a cikin kowane takamaiman rikodin a cikin fakitin da ake bincika don kamance. Wato, bayan sarrafa sabon fakitin fakitin bayanan jihar famfo tare da gwajin ƙididdiga, muna da damar (a) ƙara wannan fakitin zuwa tsarin horo don horar da ƙirar AI/ML da (b) gudanar da ingantaccen sarrafa kayan aikin. sigar yanzu na samfurin lokacin amfani da shi zuwa wannan fakitin.

Hoto 6 Aiwatar da samfurin koyo na inji zuwa bayanai masu shigowa a kan bangon “lakabin” lahani.

A daya daga cikin abubuwan da muka gabata webinars Mun nuna da kuma bayyana yadda dandalin InterSystems IRIS ya ba ku damar aiwatar da kowane tsarin AI / ML a cikin hanyar ci gaba da aiwatar da ayyukan kasuwanci wanda ke kula da amincin sakamakon ƙira da daidaita sigogin samfuri. Lokacin aiwatar da samfurin yanayin yanayin mu tare da famfo, muna amfani da duk ayyukan InterSystems IRIS da aka gabatar yayin webinar - aiwatarwa a cikin tsarin nazari a matsayin wani ɓangare na maganin mu ba koyo na gargajiya ba, amma ƙarfafa koyo, wanda ke sarrafa zaɓi ta atomatik don samfuran horo. . Samfurin horon ya ƙunshi bayanan da aka samu "ijma'i na ganowa" bayan amfani da duka gwajin ƙididdiga da sigar ƙirar yanzu - watau duka gwajin ƙididdiga (bayan canza ma'anar kamanni zuwa 0 ko 1) kuma samfurin ya haifar da sakamakon. a kan irin waɗannan bayanan 1. A lokacin sabon horo na samfurin, a lokacin tabbatarwa (sabon samfurin horarwa yana amfani da samfurin horo na kansa, tare da aikace-aikacen farko na gwajin ƙididdiga zuwa gare shi), bayanan da "bai riƙe" sakamakon 1 ba bayan aiki. ta hanyar gwajin ƙididdiga (saboda ci gaba da kasancewa a cikin horarwa samfurin rikodin daga ainihin "lakabin" na lahani), an cire shi daga tsarin horo, kuma sabon samfurin samfurin ya koya daga "lakabin" na lahani tare da bayanan "cire" daga rafi.

Hoto 7 Robotization na AI/ML lissafin a InterSystems IRIS

Idan akwai buƙatar wani nau'in "ra'ayi na biyu" game da ingancin ganowa da aka samu yayin lissafin gida a cikin InterSystems IRIS, an ƙirƙiri wani tsari mai ba da shawara don yin horo da aikace-aikacen samfura akan saitin sarrafa bayanai ta amfani da sabis na girgije (misali, Microsoft). Azure, Amazon Web Services, Google Cloud Platform, da dai sauransu):

Hoto 8 Ra'ayi na Biyu daga Microsoft Azure wanda InterSystems IRIS ta tsara

Samfurin yanayin mu a cikin InterSystems IRIS an tsara shi azaman tsarin tushen tushen tsarin hanyoyin bincike waɗanda ke hulɗa tare da kayan kayan aiki (famfo), yanayin ƙirar lissafi (Python, R da Julia), da kuma tabbatar da koyan kai na duk AI / Hanyoyin ML - akan rafukan bayanai na ainihin lokaci.

Hoto 9 Babban ayyuka na ainihin-lokaci AI/ML bayani a cikin InterSystems IRIS

Sakamakon aiki na samfurin mu:

• Samfurin lahani wanda samfurin ya gane (Janairu 12):

• Wani lahani mai tasowa wanda samfurin ya gane wanda ba a haɗa shi a cikin samfurin ba (11 ga Satumba, ƙungiyar gyara ta gano lalacewar kanta bayan kwana biyu kawai, ranar 13 ga Satumba):

Kwaikwayo a kan ainihin bayanan da ke dauke da abubuwa da yawa na lahani iri ɗaya ya nuna cewa maganinmu, wanda aka aiwatar a kan dandalin InterSystems IRIS, yana ba mu damar gano ci gaban lahani na wannan nau'in kwanaki da yawa kafin a gano su ta hanyar gyaran gyare-gyare.

InterSystems IRIS - dandamalin kwamfuta na AI/ML na ainihi na duniya

InterSystems IRIS dandamali yana sauƙaƙa haɓakawa, ƙaddamarwa da aiki na hanyoyin magance bayanai na lokaci-lokaci. InterSystems IRIS yana da ikon aiwatar da ma'amala da sarrafa bayanai a lokaci guda; goyan bayan ra'ayoyin bayanan aiki tare bisa ga ƙira da yawa (ciki har da alaƙa, matsayi, abu da takarda); yi aiki azaman dandamali don haɗa nau'ikan tushen bayanai da aikace-aikacen mutum ɗaya; samar da nazarce-nazarcen ci-gaba na ainihin-lokaci akan tsararru da bayanan da ba a tsara su ba. InterSystems IRIS kuma yana ba da hanyoyin yin amfani da kayan aikin nazari na waje kuma yana ba da damar haɗaɗɗiyar haɗin kai a cikin gajimare da kan sabar gida.

Aikace-aikacen da aka gina akan dandamali na InterSystems IRIS ana watsa su a cikin masana'antu daban-daban, suna taimaka wa kamfanoni su fahimci fa'idodin tattalin arziƙi mai mahimmanci daga hangen nesa da aiki, haɓaka ingantaccen yanke shawara da daidaita rarrabuwa tsakanin taron, bincike da aiki.

Hoto 10 InterSystems IRIS gine a cikin mahallin AI/ML na ainihi

Kamar zane na baya, zanen da ke ƙasa ya haɗu da sabon "tsarin daidaitawa" (CD / CI / CT) tare da zane na kwararar bayanai tsakanin abubuwan aiki na dandamali. Nunin gani yana farawa da CD ɗin macromechanism kuma yana ci gaba da macromechanisms CI da CT.

Hoto 11 Tsarin bayanai yana gudana tsakanin abubuwan AI/ML na dandalin InterSystems IRIS

Ma'anar tsarin CD a cikin InterSystems IRIS: masu amfani da dandamali (masu haɓaka AI / ML mafita) daidaitawa da / ko ƙirƙirar sabon ci gaban AI / ML ta amfani da editan lambar musamman don hanyoyin AI / ML: Jupyter (cikakken suna: Jupyter Notebook; a takaice, takardun da aka ƙirƙira a cikin wannan editan ma wasu lokuta ana kiran su). A cikin Jupyter, mai haɓakawa yana da damar rubutawa, gyarawa da kuma tabbatar da aikin (ciki har da amfani da zane-zane) na takamaiman ci gaban AI / ML kafin a sanya shi ("an tura") a cikin InterSystems IRIS. A bayyane yake cewa sabon ci gaba da aka kirkira ta wannan hanyar zai sami kawai kuskuren asali (tunda, musamman, Jupyter baya aiki tare da rafukan bayanan lokaci-lokaci) - wannan yana cikin tsari na abubuwa, saboda babban sakamakon ci gaba a Jupyter tabbaci ne na ainihin aiki na tsarin AI / ML daban ("yana nuna sakamakon da ake tsammani akan samfurin bayanai"). Hakazalika, tsarin da aka riga aka sanya shi a cikin dandamali (duba waɗannan macro-machanisms masu zuwa) kafin gyarawa a Jupyter na iya buƙatar "juyawa" zuwa nau'i na "pre-platform" (karanta bayanai daga fayiloli, aiki tare da bayanai ta hanyar xDBC maimakon tebur, hulɗa kai tsaye tare da duniya - multidimensional data arrays InterSystems IRIS - da dai sauransu).

Wani muhimmin al'amari na aiwatar da CD a cikin InterSystems IRIS: an aiwatar da haɗin kai tsakanin dandamali da Jupyter, ƙyale abun ciki a cikin Python, R da Julia don canjawa wuri zuwa dandamali (kuma, daga baya, ana sarrafa su a cikin dandamali) (duk ukun suna shirye-shirye). Harsuna a cikin manyan harsunan buɗe tushen tushen). Don haka, masu haɓaka abun ciki na AI / ML suna da damar aiwatar da "ci gaba da turawa" wannan abun cikin dandamali, suna aiki a cikin editan Jupyter da suka saba, tare da sanannun ɗakunan karatu da ke cikin Python, R, Julia, da yin gyara na asali (idan ya cancanta) a wajen dandalin .

Bari mu matsa zuwa tsarin macro na CI a cikin InterSystems IRIS. Jadawalin yana nuna tsarin macro na "robotizer na ainihi" (haɗin tsarin bayanai, hanyoyin kasuwanci da gutsuttsuran lambobi waɗanda aka tsara su a cikin harsunan lissafi da ObjectScript - harshen ci gaban ƙasa na InterSystems IRIS). Ayyukan wannan tsari na macro shine kula da layukan bayanan da suka wajaba don aiwatar da hanyoyin AI / ML (dangane da rafukan bayanan da aka watsa zuwa dandamali a cikin ainihin lokacin), yanke shawara game da jerin aikace-aikacen da "haɗin kai" na AI / Hanyoyin ML (su kuma "algorithms na lissafi", "samfurin", da dai sauransu - ana iya kiran su daban-daban dangane da ƙayyadaddun abubuwan aiwatarwa da abubuwan da ake so), kiyaye tsarin bayanai har zuwa yau don nazarin sakamakon aikin AI / Hanyoyin ML (cubes, Tables, multidimensional data arrays, da dai sauransu) da sauransu - don rahotanni, dashboards, da dai sauransu).

Wani muhimmin al'amari na aiwatar da CI musamman a cikin InterSystems IRIS: an aiwatar da haɗin kai tsakanin dandamali da yanayin ƙirar lissafi, yana ba ku damar aiwatar da abubuwan da aka shirya a dandamali a cikin Python, R da Julia a cikin mahallin su kuma ku sami dawo da aiwatar da aiwatarwa. sakamako. Ana aiwatar da wannan haɗin kai duka biyu a cikin "yanayin tasha" (watau, AI / ML abun ciki an tsara shi azaman lambar ObjectScript wanda ke yin kira zuwa yanayin) kuma a cikin "yanayin tsarin kasuwanci" (watau, AI / ML abun ciki an tsara shi azaman tsarin kasuwanci). ta amfani da editan hoto, ko wani lokacin amfani da Jupyter, ko amfani da IDE - IRIS Studio, Eclipse, Visual Studio Code). Samun hanyoyin kasuwanci don gyarawa a Jupyter yana nunawa ta hanyar haɗin kai tsakanin IRIS a matakin CI da Jupyter a matakin CD. An ba da ƙarin cikakken bayyani na haɗin kai tare da yanayin ƙirar lissafi a ƙasa. A wannan mataki, a cikin ra'ayinmu, akwai kowane dalili don tabbatar da cewa dandamali yana da duk kayan aikin da ake bukata don aiwatar da "ci gaba da haɗin kai" na ci gaba na AI / ML (wanda ke fitowa daga "ci gaba da ƙaddamarwa") a cikin ainihin AI / ML mafita.

Kuma babban tsarin macro: CT. Idan ba tare da shi ba, babu wani dandamali na AI / ML (ko da yake "ainihin lokacin" za a aiwatar ta hanyar CD / CI). Mahimmancin CT shine aikin dandamali tare da "kayan kayan tarihi" na koyo na inji da kuma basirar wucin gadi kai tsaye a cikin zaman aiki na yanayin ƙirar lissafi: samfuri, teburin rarrabawa, matrix vectors, layers of neuro networks, da dai sauransu. Wannan "aiki", a mafi yawan lokuta, ya ƙunshi ƙirƙirar abubuwan da aka ambata a cikin mahalli (a cikin yanayin samfuri, alal misali, "halitta" ya ƙunshi saita ƙayyadaddun samfurin da zaɓi na gaba na ƙimar sigoginsa - abin da ake kira "horo" na samfurin), aikace-aikacen su (don samfurori: ƙididdiga tare da taimakon su na "samfurin" dabi'u na masu canji - kisa, membobin rukuni, yiwuwar wani taron, da dai sauransu) da kuma ingantawa na riga. kayan tarihi da aka ƙirƙira da amfani da su (misali, sake fasalin saiti na sauye-sauyen shigar da ƙima dangane da sakamakon aikace-aikacen - don haɓaka daidaiton tsinkaya, a matsayin zaɓi). Mahimmin mahimmancin fahimtar muhimmancin CT shine "abstraction" daga ainihin CD da CI: CT zai aiwatar da duk kayan tarihi, yana mai da hankali kan ƙayyadaddun ƙididdiga da lissafi na AI / ML bayani a cikin iyawar da aka bayar ta wasu wurare. Alhakin "samar da bayanai" da "samar da kayan aiki" zai kasance alhakin CD da CI.

Wani muhimmin al'amari na aiwatar da CT musamman a cikin InterSystems IRIS: ta yin amfani da haɗin kai tare da yanayin ƙirar lissafin lissafi da aka riga aka ambata a sama, dandamali yana da ikon cire waɗannan kayan tarihi na musamman daga zaman aikin da ke gudana a ƙarƙashin ikonsa a cikin yanayin lissafi kuma (mafi mahimmanci) juya. su cikin abubuwan dandali data. Misali, teburin rarraba da aka ƙirƙira a cikin zaman Python mai aiki zai iya zama (ba tare da dakatar da zaman Python ba) a canza shi zuwa dandamali a cikin nau'in, alal misali, na duniya (tsararrun bayanan InterSystems IRIS multidimensional) - kuma ana amfani da su. don ƙididdigewa a cikin wani tsarin AI/ML (wanda aka aiwatar a cikin yaren wani yanayi - alal misali, a cikin R) - ko tebur mai kama-da-wane. Wani misali: a cikin layi daya tare da "yanayin al'ada" na aikin samfurin (a cikin aikin aiki na Python), "auto-ML" ana aiwatar da shi akan bayanan shigar da shi: zaɓin atomatik na madaidaitan ma'auni na shigarwa da ƙimar sigina. Kuma tare da horo na "na yau da kullum", samfurin mai inganci a cikin ainihin lokaci kuma yana karɓar "shawarwari don ingantawa" na ƙayyadaddun sa - wanda saitin masu canji na shigarwa ya canza, ma'auni suna canzawa (ba a sakamakon horo ba. a cikin Python, amma sakamakon horarwa tare da sigar “madadin” na kanta, kamar tari na H2O), yana ba da damar gabaɗayan AI / ML mafita don jure sauye-sauyen da ba zato ba tsammani a yanayin shigar bayanan da abubuwan abubuwan da ake ƙira. .

Bari mu san daki-daki tare da aikin AI/ML na InterSystems IRIS, ta amfani da misalin samfurin rayuwa ta gaske.

A cikin zanen da ke ƙasa, a gefen hagu na faifan akwai wani ɓangare na tsarin kasuwanci wanda ke aiwatar da aiwatar da rubutun a cikin Python da R. A tsakiyar ɓangaren akwai na'urorin gani na aiwatar da wasu daga cikin waɗannan rubutun, bi da bi. a cikin Python da R. Nan da nan a bayansu akwai misalan abubuwan da ke cikin ɗaya da wani harshe, an canza su don aiwatarwa zuwa wuraren da suka dace. A ƙarshe a hannun dama akwai abubuwan gani bisa sakamakon aiwatar da rubutun. Abubuwan da aka gani a saman an yi su ne a kan IRIS Analytics (an samo bayanai daga Python zuwa cikin InterSystems IRIS data dandamali kuma an nuna su a kan dashboard ta amfani da dandamali), a ƙasa an yi su kai tsaye a cikin zaman R aiki da fitarwa daga can zuwa fayilolin hoto. . Wani muhimmin al'amari: guntun da aka gabatar a cikin samfurin yana da alhakin horar da samfurin (rarrabuwar jihohin kayan aiki) akan bayanan da aka karɓa a ainihin lokacin daga tsarin na'urar kwaikwayo na kayan aiki, bisa umarni daga tsarin kula da ingancin ƙira da aka lura yayin aikace-aikacen samfurin. Za a ci gaba da yin magana game da aiwatar da wani bayani na AI / ML a cikin nau'i na tsarin tsarin hulɗa ("wakilai").

Hoto 12 Yin hulɗa tare da Python, R da Julia a cikin InterSystems IRIS

Tsarin dandamali (su ma "hanyoyin kasuwanci", "hanyoyi na nazari", "bututun", da dai sauransu - dangane da mahallin), da farko, ana gyara su a cikin editan tsarin kasuwanci na hoto a cikin dandalin kanta, kuma a cikin irin wannan tsarin. yadda za a ƙirƙiri zane-zanen toshewar sa da kuma tsarin AI/ML daidai (lambar shirin) a lokaci guda. Lokacin da muka ce "ana samun hanyar AI / ML," da farko muna nufin haɓakawa (a cikin tsari ɗaya): abun ciki a cikin harsunan yanayin ƙirar lissafi yana kusa da abun ciki a cikin SQL (ciki har da kari daga HadakarML), a cikin InterSystems ObjectScript, tare da wasu harsuna masu tallafi. Haka kuma, tsarin dandamali yana ba da dama mai fa'ida don "sadarwa" a cikin nau'i na ɓarke ​​​​tsayi (kamar yadda ake iya gani a cikin misali a cikin zanen da ke ƙasa), wanda ke ba ku damar tsara yadda ya kamata ko da hadaddun abun ciki ba tare da taɓa “faɗuwa ba” na tsarin zane (zuwa tsarin "marasa hoto"). »hanyoyi / azuzuwan / tsari, da sauransu). Wato, idan ya cancanta (kuma ana iya hango shi a yawancin ayyukan), gaba ɗaya za a iya aiwatar da duk maganin AI / ML a cikin sigar rubutun kai na hoto. Da fatan za a lura cewa a cikin tsakiyar ɓangaren zanen da ke ƙasa, wanda ke wakiltar babban "matakin gida", ya bayyana a fili cewa ban da ainihin aikin horar da samfurin (ta amfani da Python da R), nazarin abin da ake kira. ROC lankwasa na horar da model aka kara, kyale gani (da lissafi ma) kimanta ingancin horo - da kuma wannan bincike da aka aiwatar a cikin Julia harshen (aka aiwatar, daidai da, a cikin Julia lissafi yanayi).

Hoto 13 Yanayin gani don abun da ke ciki na AI / ML mafita a cikin InterSystems IRIS

Kamar yadda aka ambata a baya, farkon haɓakawa da (a wasu lokuta) daidaitawar hanyoyin AI / ML da aka riga aka aiwatar a cikin dandamali za a iya yin su a waje da dandamali a cikin editan Jupyter. A cikin zanen da ke ƙasa mun ga misali na daidaita tsarin dandali da ake da su (kamar yadda yake a cikin zanen da ke sama) - wannan shine yadda guntun da ke da alhakin horar da samfurin yayi kama da Jupyter. Ana samun abun ciki na Python don gyarawa, gyarawa, da fitar da zane kai tsaye a cikin Jupyter. Ana iya yin canje-canje (idan ya cancanta) tare da aiki tare nan take cikin tsarin dandamali, gami da sigar sa mai amfani. Ana iya canja wurin sabon abun ciki zuwa dandamali ta hanya mai kama da (sabon tsarin dandamali yana haifar da shi ta atomatik).

Hoto 14 Yin amfani da littafin rubutu na Jupyter don gyara injin AI/ML a cikin tsarin InterSystems IRIS

Ana iya daidaita tsarin dandamali ba kawai a cikin hoto ko tsarin kwamfutar tafi-da-gidanka ba - har ma a cikin tsarin “jimla” IDE (Integrated Development Environment). Waɗannan IDEs sune IRIS Studio (situdiyon IRIS na asali), Kayayyakin Kayayyakin Kayayyakin Kayayyakin (InterSystems IRIS tsawo don VSCode) da Eclipse (Atelier plugin). A wasu lokuta, yana yiwuwa ƙungiyar ci gaba ta yi amfani da duk IDE guda uku a lokaci guda. Hoton da ke ƙasa yana nuna misali na gyara tsari iri ɗaya a cikin ɗakin studio na IRIS, a cikin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Kayayyakin Hulda) da kuma a cikin Eclipse. Babu shakka duk abun ciki yana samuwa don gyarawa: Python/R/Julia/SQL, ObjectScript, da tsarin kasuwanci.

Hoto 15 Ci gaban InterSystems tsarin kasuwanci na IRIS a cikin IDE daban-daban

Kayan aikin don bayyanawa da aiwatar da hanyoyin kasuwanci na InterSystems IRIS a cikin Harshen Tsarin Kasuwanci (BPL) sun cancanci ambaton musamman. BPL yana ba da damar yin amfani da "ayyukan haɗin kai da aka shirya" (ayyukan) a cikin ayyukan kasuwanci - wanda, a zahiri, yana ba da kowane dalili na faɗin cewa ana aiwatar da "ci gaba da haɗin kai" a cikin InterSystems IRIS. Shirye-shiryen tsarin kasuwancin da aka ƙera (ayyukan da haɗin kai tsakanin su) babban haɓakawa ne don haɗa maganin AI / ML. Kuma ba kawai majalisai ba: godiya ga ayyukan da haɗin kai a tsakanin su a kan ci gaban AI / ML daban-daban da kuma hanyoyin, "launi mai sarrafa kansa" ya tashi, yana iya yanke shawara bisa ga halin da ake ciki, a ainihin lokacin.

Hoto 16 Shirye-shiryen tsarin kasuwancin da aka yi don ci gaba da haɗin kai (CI) akan dandalin InterSystems IRIS

Manufar tsarin wakilai (wanda kuma aka sani da "tsarin wakilai da yawa") yana da matsayi mai ƙarfi a cikin aikin mutum-mutumi, kuma tsarin InterSystems IRIS yana goyan bayan shi ta hanyar gina "samfurin-tsari". Baya ga damar da ba ta da iyaka don "kaya" kowane tsari tare da ayyukan da suka wajaba don cikakken bayani, ba da tsarin tsarin dandamali tare da mallakar "Hukumar" yana ba ku damar ƙirƙirar ingantattun mafita don abubuwan da ba a iya daidaita su ba (halayen zamantakewa / halin zamantakewa. biosystems, partially observable fasaha tafiyar matakai, da dai sauransu).

Hoto 16 Aiki na maganin AI / ML azaman tsarin tsarin kasuwanci na tushen wakili a cikin InterSystems IRIS

Muna ci gaba da bitar InterSystems IRIS tare da labari game da amfani da dandamali don magance duka azuzuwan matsalolin lokaci-lokaci (cikakkiyar gabatarwar ga wasu mafi kyawun ayyuka na dandamali AI / ML akan InterSystems IRIS za a iya samu a ɗaya. na mu na baya webinars).

Hot a kan diddige na zane na baya, a ƙasa akwai ƙarin cikakkun bayanai na tsarin wakili. Shafin yana nuna alamun iri ɗaya, duk hanyoyin da ke tsakaninsu suna tsari ne na kayan aikin da ke cikin kanta, saka idanu - suna aiwatar da injin da kansu, saka idanu ingancin koyo na inji da kuma ƙaddamar da sigina game da buƙatar sake horar da samfurin.

Hoto 17 Haɗawar maganin AI / ML a cikin tsarin tsarin kasuwanci na tushen wakili a cikin InterSystems IRIS

Hoton da ke ƙasa yana kwatanta aikin kansa na wani samfurin mutum-mutumi (gane da canza launin rubutu) na ɗan lokaci. A cikin babba shine juyin halitta na ingantacciyar alamar horon ƙirar (ingancin yana haɓakawa), a cikin ƙananan ɓangaren shine haɓakar ƙimar ingancin aikace-aikacen ƙirar da kuma gaskiyar maimaita horo (jajayen ratsi). Kamar yadda kake gani, maganin ya koyi kanta da kyau kuma mai zaman kansa, kuma yana aiki a matakin ingancin da aka ba shi (ƙimar ƙimar ƙimar ba ta faɗi ƙasa da 80%).

Hoto 18 Ci gaba da horar da kai (CT) akan dandalin InterSystems IRIS

Mun kuma ambata "auto-ML" a baya, amma zanen da ke ƙasa yana nuna amfani da wannan aikin daki-daki ta amfani da misalin wani samfuri. Hoton zane na guntun tsarin kasuwanci yana nuna ayyukan da ke haifar da yin samfuri a cikin tarin H2O, yana nuna sakamakon wannan ƙirar (mafi girman rinjayen samfurin da aka samu akan ƙirar “wanda mutum ya yi”), bisa ga kwatancen kwatancen. ROC masu lankwasa, da kuma ganowa ta atomatik na "mafi tasiri masu tasiri" da ke cikin saitin bayanan asali). Wani muhimmin batu a nan shi ne adana lokaci da albarkatun ƙwararru waɗanda aka samu ta hanyar "auto-ML": abin da tsarin dandalinmu ya yi a cikin rabin minti (nema da horar da mafi kyawun samfurin) zai iya ɗaukar gwani daga mako guda zuwa wata.

Hoto 19 Haɗuwa da "auto-ML" a cikin maganin AI / ML akan dandalin InterSystems IRIS

Hoton da ke ƙasa ya rasa ma'anar kadan, amma hanya ce mai kyau don kawo karshen labarin game da azuzuwan matsalolin da ake magance matsalolin lokaci-lokaci: muna tunatar da ku cewa tare da duk damar da tsarin InterSystems IRIS, tsarin horarwa a ƙarƙashin ikonsa shine. ba wajibi ba. Dandalin zai iya karɓa daga waje abin da ake kira PMML ƙayyadaddun samfurin, wanda aka horar da shi a cikin kayan aiki wanda ba a karkashin kulawar dandamali ba - kuma yana amfani da wannan samfurin a ainihin lokacin daga lokacin da aka shigo da shi. Bayanin PMML. Yana da mahimmanci a la'akari da cewa ba duk kayan aikin AI / ML ba ne za a iya rage su zuwa ƙayyadaddun PMML, ko da mafi yawan kayan tarihi na yau da kullum sun ba da damar wannan. Don haka, dandalin InterSystems IRIS shine "bude madauki" kuma baya nufin "bautar dandamali" ga masu amfani.

Hoto 20 Haɗuwa da "auto-ML" a cikin maganin AI / ML akan dandalin InterSystems IRIS

Bari mu lissafa ƙarin fa'idodin dandamali na InterSystems IRIS (don tsabta, dangane da sarrafa sarrafawa), waɗanda ke da matukar mahimmanci a cikin sarrafa kai na haƙƙin ɗan adam da koyan injin na ainihi:

• Haɓaka kayan aikin haɗin kai tare da kowane tushen bayanai da masu amfani (tsarin sarrafa tsari/SCADA, kayan aiki, MRO, ERP, da sauransu)
• Ginannen Multi-model DBMS don babban aiki ma'amala da sarrafa nazari (Hybrid Transaction/Analytical Processing, HTAP) na kowane girma na bayanan aiwatar da fasaha
• Kayan aikin haɓaka don ci gaba da tura injin AI / ML don mafita na ainihin lokaci dangane da Python, R, Julia
• Hanyoyin kasuwanci masu dacewa don ci gaba da haɗin kai da (kai-) koyo na ainihin lokaci AI/ML mafita
• Gina kayan aikin Intelligence na Kasuwanci don ganin bayanan tsari da sakamakon maganin AI/ML
• Gudanarwar API don isar da sakamakon AI / ML bayani don aiwatar da tsarin sarrafawa / SCADA, bayanai da tsarin nazari, aika faɗakarwa, da dai sauransu.

Hanyoyin AI / ML akan dandamalin InterSystems IRIS cikin sauƙi sun dace da abubuwan da ke akwai na IT. InterSystems IRIS dandamali yana tabbatar da babban amincin mafita na AI / ML ta hanyar goyan bayan ƙayyadaddun juzu'i da bala'i da ƙayyadaddun ƙayyadaddun aiki a cikin mahalli mai kama-da-wane, akan sabobin jiki, a cikin girgije masu zaman kansu da na jama'a, da kwantena Docker.

Don haka, InterSystems IRIS dandamali ne na ainihin lokacin AI/ML na duniya. An tabbatar da duniyar duniyar dandalinmu a aikace ta hanyar rashin haƙƙin haƙƙin haƙƙin haƙƙin haƙƙin haƙƙin haƙƙin haƙƙin ƙididdiga da aka aiwatar, ikon InterSystems IRIS don haɗawa (a ainihin lokacin) sarrafa al'amuran daga masana'antu iri-iri, da kuma daidaitawa na musamman. kowane dandali ayyuka da hanyoyin zuwa takamaiman bukatun masu amfani.

Hoto 21 InterSystems IRIS - dandamalin kwamfuta na AI/ML na ainihi na duniya

Don ƙarin ma'amala mai mahimmanci tare da na masu karatunmu waɗanda ke sha'awar abubuwan da aka gabatar anan, muna ba da shawarar kada ku iyakance kanku ga karanta shi kuma ku ci gaba da tattaunawa "rayuwa." Za mu yi farin cikin ba da tallafi tare da ƙirƙirar yanayin AI / ML na ainihi dangane da ƙayyadaddun kamfanin ku, aiwatar da samfuran haɗin gwiwa akan dandamalin InterSystems IRIS, ƙirƙira da aiwatarwa a aikace taswirar hanya don gabatar da hankali na wucin gadi da koyon injin. cikin ayyukan samarwa da sarrafa ku. Imel ɗin Tuntuɓar Ƙwararrun Ƙwararrun Ƙwararrun mu na AI/ML - [email kariya].

source: www.habr.com