IBM e bula CodeNet bakeng sa litsamaiso tsa ho ithuta tsa mochini tse fetolelang le ho netefatsa khoutu

IBM e senotse morero oa eona oa CodeNet, o ikemiselitseng ho fa bafuputsi dataset e ba lumellang hore ba leke tšebeliso ea mekhoa ea ho ithuta ka mochine ho etsa bafetoleli ba lipuo tsa mananeo, lijenereithara tsa khoutu le bahlahlobisisi. CodeNet e kenyelletsa pokello ea mehlala ea likhoutu tse limilione tse 14 tse rarollang mathata a 4053 a tloaelehileng a mananeo. Ka kakaretso, pokello e na le mela e ka bang limilione tse 500 ea khoutu mme e akaretsa lipuo tse 55 tsa lenaneo, lipuo tsa sejoale-joale tse kang C ++, Java, Python le Go, le lipuo tsa lefa tse kenyelletsang COBOL, Pascal le FORTRAN. Lintlafatso tsa morero li ajoa tlasa laesense ea Apache 2.0, 'me lisebelisoa tsa data li reretsoe ho ajoa ka mokhoa oa sechaba.

Mehlala e hlalositsoe 'me e sebelisa li-algorithms tse tšoanang lipuong tse fapaneng tsa lenaneo. Ho lebeletsoe hore sete e reriloeng e tla thusa ho koetlisa mekhoa ea ho ithuta mochine le ho hlahisa mekhoa e mecha tšimong ea phetolelo le mochine oa khoutu ea mochine, ho tšoana le kamoo setšoantšo sa setšoantšo sa ImageNet se hlalositsoeng se thusitseng nts'etsopele ea ho lemoha mohlala le mekhoa ea pono ea k'homphieutha. Litlholisano tse fapaneng tsa mananeo li boleloa e le o mong oa mehloli ea mantlha ea ho theha pokello.

Ho fapana le bafetoleli ba setso, ba sebelisoang ho latela melao ea phetoho, mekhoa ea ho ithuta ka mochine e ka tšoara le ho nahanela moelelo oa tšebeliso ea khoutu. Ha u fetolela ho tloha puong e 'ngoe ea lenaneo ho ea ho e' ngoe, moelelo o bohlokoa joalo ka ha o fetolela ho tloha puong e 'ngoe ea batho ho ea ho e' ngoe. Ho haella hona ha ho nahanela maemo ke hona ho thibelang khoutu ho fetoloa ho tsoa lipuong tse tsebahalang joalo ka COBOL.

Ho ba le database e kholo ea ts'ebetsong ea algorithm ka lipuo tse fapaneng ho tla thusa ho theha lits'ebetso tsa ho ithuta tsa mochini tsa bokahohleng tseo, ho fapana le phetolelo e tobileng lipakeng tsa lipuo tse itseng, li laolang kemiso e hlakileng haholoanyane ea khoutu, e ikemetseng ka lipuo tse itseng tsa lenaneo. Sistimi e joalo e ka sebelisoa e le mofetoleli, e fetolela khoutu e fetisitsoeng ka puo efe kapa efe e tšehelitsoeng ka har'a boemeli ba eona ba kahare, moo khoutu e ka hlahisoang ka lipuo tse ngata.

Sistimi e tla boela e khone ho etsa liphetoho tse fapaneng. Ka mohlala, libanka le mekhatlo ea 'muso e ntse e tsoela pele ho sebelisa merero ka puo ea COBOL ea khale. Mofetoleli ea thehiloeng mochineng oa ho ithuta o tla khona ho fetola khoutu ea COBOL hore e be setšoantšo sa Java, 'me, ha ho hlokahala, a fetolele sekhechana sa Java ho khoutu ea COBOL.

Ntle le ho fetolela lipakeng tsa lipuo, likarolo tse joalo tsa ts'ebeliso ea CodeNet joalo ka ho theha lits'ebetso tse bohlale tsa ho batla khoutu le boiketsetso ba ho lemoha ha clone, hammoho le nts'etsopele ea li-optimizers le litsamaiso tsa tokiso ea khoutu li boletsoe. Haholo-holo, mehlala e hlahisitsoeng ho CodeNet e na le metadata e hlalosang liphetho tsa tlhahlobo ea ts'ebetso, boholo ba lenaneo le hlahisoang, tšebeliso ea mohopolo le naha, e re lumellang ho khetholla khoutu e nepahetseng ho khoutu e nang le liphoso (ho khetholla khoutu e nepahetseng ho khoutu e fosahetseng, pokello ka ho khetheha e kenyelletsa mehlala e nang le liphoso, karolo ea eona e leng 29.5%). Mokhoa oa ho ithuta oa mochine o ka nka metadata ena ho hlahisa khoutu e nepahetseng ka ho fetisisa kapa ho tseba ho fokotseha ho khoutu e hlahlobiloeng (sistimi e ka utloisisa hore algorithm khoutu e rometsoeng ha e sebelisoe hantle kapa e na le liphoso).

Source: opennet.ru

Eketsa ka tlhaloso