IBM mepere CodeNet maka sistemu mmụta igwe na-atụgharị ma chọpụta koodu

IBM ewepụtala atụmatụ CodeNet ya, nke na-achọ ịnye ndị nyocha dataset nke na-enye ha ohere ịnwale iji usoro mmụta igwe mepụta ndị ntụgharị asụsụ mmemme, ndị na-emepụta koodu na ndị nyocha. CodeNet gụnyere nchịkọta koodu koodu nde 14 na-edozi nsogbu mmemme 4053. Na mkpokọta, mkpokọta ahụ nwere ihe dị ka ahịrị koodu nde 500 ma kpuchie asụsụ mmemme 55, ma asụsụ ọgbara ọhụrụ dị ka C ++, Java, Python na Go, yana asụsụ ndị nketa gụnyere COBOL, Pascal na FORTRAN. A na-ekesa mmepe nke oru ngo a n'okpuru ikike Apache 2.0, na a na-eme atụmatụ ikesa data n'ụdị nke ọha.

A na-akọwapụta ihe atụ ndị a ma mejuputa algọridim ndị ahụ n'asụsụ mmemme dị iche iche. A na-atụ anya na ihe a na-atụ anya ga-enyere aka ịzụ usoro mmụta igwe na ịmepụta ihe ọhụrụ na ngalaba ntụgharị asụsụ na nhazi koodu igwe, yiri ka ImageNet nke nkọwapụta ihe oyiyi si nyere aka n'ichepụta ụkpụrụ na usoro ọhụụ kọmputa. Asọmpi mmemme dị iche iche ka a na-akpọ dịka otu n'ime isi mmalite nke nchịkọta.

N'adịghị ka ndị ntụgharị ọdịnala, nke a na-emejuputa atumatu dabere na iwu ntụgharị asụsụ, usoro mmụta igwe nwere ike ijide ma buru n'uche ọnọdụ ojiji koodu. Mgbe ị na-atụgharị site n'otu asụsụ mmemme gaa na nke ọzọ, ihe gbara ya gburugburu dị mkpa ka mgbe ị na-atụgharị site n'otu asụsụ mmadụ gaa na nke ọzọ. Enweghị nlebara anya gburugburu ebe a bụ ihe na-egbochi koodu ịtụgharị site na asụsụ ndị nwere aha dị ka COBOL.

Inwe nnukwu nchekwa data nke mmejuputa algọridim n'asụsụ dị iche iche ga-enyere aka ịmepụta usoro mmụta igwe zuru ụwa ọnụ nke, kama ịtụgharị asụsụ ozugbo n'etiti asụsụ dị iche iche, na-emegharị ihe nnọchianya nke koodu ahụ, na-adabereghị na asụsụ mmemme. Enwere ike iji usoro dị otú ahụ dị ka onye ntụgharị, na-atụgharị koodu a na-ekesa n'asụsụ ọ bụla akwadoro ka ọ bụrụ ihe nnọchianya nke ime ya, nke enwere ike ịmepụta koodu n'ọtụtụ asụsụ.

Usoro a ga-enwekwa ike ime mgbanwe ngbanwe bidirectional. Dịka ọmụmaatụ, ụlọ akụ na ụlọ ọrụ gọọmentị na-aga n'ihu na-eji ọrụ eme ihe n'asụsụ COBOL oge ochie. Onye ntụgharị asụsụ dabere na mmụta igwe ga-enwe ike ịtụgharị koodu COBOL ka ọ bụrụ nnọchite Java, ma, ọ bụrụ na ọ dị mkpa, tụgharịa akụkụ Java azụ ka ọ bụrụ koodu COBOL.

Na mgbakwunye na ntụgharị n'etiti asụsụ, a na-akpọ mpaghara ngwa CodeNet dị ka imepụta sistemụ nchọ koodu smart na akpaaka nke nchọpụta clone, yana mmepe nke optimizers na sistemụ mgbazi koodu akpaka. Karịsịa, ihe atụ ndị a gosipụtara na CodeNet nwere metadata na-akọwa nsonaazụ nke nnwale arụmọrụ, nha nke mmemme, oriri ebe nchekwa na steeti, nke na-enye anyị ohere ịmata ọdịiche dị na koodu ziri ezi na koodu nwere njehie (iji mata ọdịiche ziri ezi na koodu na-ezighi ezi, nchịkọta gụnyere kpọmkwem ihe atụ nwere njehie, nke òkè ya bụ 29.5%). Usoro mmụta igwe nwere ike iburu metadata a n'uche iji mepụta koodu kachasị mma ma ọ bụ iji chọpụta regressions na koodu nyochara (usoro ahụ nwere ike ịghọta na algọridim dị na koodu e debere adịghị arụ ọrụ nke ọma ma ọ bụ nwee njehie).

isi: opennet.ru

Tinye a comment