GitHub yatsegula chitukuko pakugwiritsa ntchito makina ophunzirira pakusaka ma code ndi kusanthula

GitHub anayambitsa kulemba CodeSearchNet, yomwe yakonza zitsanzo zamakina ophunzirira ndi ma data ofunikira pakugawa, kugawa ndi kusanthula ma code m'zilankhulo zosiyanasiyana zamapulogalamu. CodeSearchNet, yofanana ndi ImageNet, imaphatikizaponso zophatikiza zambiri za ma code okhala ndi mawu ofotokozera zomwe ma code amachita. Zopangira zophunzitsira ndi zitsanzo zogwiritsa ntchito CodeSearchNet zalembedwa mu Python pogwiritsa ntchito Tensorflow chimango ndi wogawidwa ndi pansi pa layisensi ya MIT.

Popanga CodeSearchNet, matekinoloje ophatikizika a zilankhulo zachilengedwe adagwiritsidwa ntchito, zomwe zimapangitsa kuti makina ophunzirira makina asamangoganizira za kaphatikizidwe, komanso tanthauzo la zomwe zimachitika ndi code. Dongosolo la GitHub kuyikidwa pakuyesa kukonza kusaka kwa ma semantic code pogwiritsa ntchito mafunso chinenero chachibadwa (Mwachitsanzo, popempha "kusanja mndandanda wa zingwe", kachidindo ndi kukhazikitsidwa kwa ma algorithms ofanana akuwonetsedwa).

Deta yomwe ikufunsidwayo ili ndi maulalo opitilira 2 miliyoni, okonzedwa kutengera zolemba zamalaibulale otseguka omwe alipo. Khodiyo imaphatikizapo zolemba zonse za ntchito kapena njira, ndipo ndemangayi ikufotokoza zomwe zimachitika ndi ntchitoyi (zolemba zatsatanetsatane zimaperekedwa). Pakadali pano, ma dataset akonzedwa Python, JavaScript, Ruby, Go, Java ndi PHP. Zitsanzo zimaperekedwa pogwiritsa ntchito ma dataset omwe akufunsidwa pophunzitsa mitundu yosiyanasiyana ya neural network, kuphatikiza Neural-Bag-Of-Mawu, RNN, Kudzisamalira (BERT) ndi 1D-CNN+Self-Attention Hybrid.

Kupanga njira zofufuzira zilankhulo zachilengedwe, gulu la CodeSearchNet Challenge lakonzedwanso, kuphatikiza
99 zofananira mafunso okhala ndi zofotokozera zaukatswiri pafupifupi 4 zofotokoza zomangirira kwambiri mu CodeSearchNet Corpus dataset, zofotokoza njira ndi ntchito pafupifupi 6 miliyoni (set size pafupifupi 20 GB). CodeSearchNet Challenge ikhoza kukhala ngati chizindikiritso chowunika momwe njira zina zofufuzira zilankhulo zachilengedwe zimagwirira ntchito. Kugwiritsa ntchito zida KubeFlow okonzeka
chitsanzo code search engine.

Source: opennet.ru

Kuwonjezera ndemanga