GitHub ua tatalaina atinaʻe i le faʻaogaina o masini aʻoaʻoga mo suʻesuʻega faʻailoga ma auʻiliʻiliga

GitHub faʻalauiloa le poloketi CodeSearchNet, lea ua saunia ai faʻataʻitaʻiga aʻoaʻoga masini ma seti faʻamaumauga e manaʻomia mo le faʻavasegaina, faʻavasegaina ma le suʻesuʻeina o code i gagana eseese polokalame. CodeSearchNet, tutusa ma IMAGEnet, e aofia ai se aofa'iga tele o snippets code ma fa'amatalaga e fa'amaonia ai mea e fai e le code. O vaega mo aʻoaʻoga faʻataʻitaʻiga ma faʻataʻitaʻiga o le faʻaogaina o le CodeSearchNet o loʻo tusia i le Python e faʻaaoga ai le Tensorflow framework ma tufatufaina e i lalo ole laisene MIT.

I le fatuina o le CodeSearchNet, na faʻaaogaina tekinolosi faʻasalalau tusitusiga, e mafai ai e masini aʻoaʻoga faʻaogaina e le gata i foliga faʻaoga, ae faʻapea foʻi le uiga o gaioiga na faia e le code. Le faiga GitHub apalai i fa'ata'ita'iga i le fa'atulagaina o su'esu'ega semantic code e fa'aaoga ai fesili i luga gagana masani (mo se faʻataʻitaʻiga, pe a talosagaina le "faʻavasegaina o se lisi o manoa", faʻailoga faʻatasi ma le faʻatinoga o algorithms tutusa e faʻaalia).

O fa'amaumauga fuafuaina e aofia ai le sili atu ma le 2 miliona feso'ota'iga code-comment, saunia e fa'atatau i tusitusiga fa'apogai o faletusi tatala oi ai nei. O le code e aofia ai le faʻamatalaga atoa o faʻamatalaga o galuega taʻitasi poʻo metotia, ma o le faʻamatalaga e faʻamatalaina ai gaioiga na faia e le galuega (o loʻo tuʻuina atu faʻamatalaga auiliili). I le taimi nei, ua saunia faʻamaumauga mo Python, JavaScript, Ruby, Go, Java ma PHP. O faʻataʻitaʻiga o loʻo tuʻuina atu i le faʻaogaina o faʻamaumauga fuafuaina mo le aʻoaʻoina o ituaiga eseese o fesoʻotaʻiga neural, e aofia ai Neural-Pe-O-Upu, RNN, Ua'i atu e le tagata lava ia (BERT) ma 1D-CNN + Fa'a'au'au Fa'atasi.

Ina ia atia'e faiga su'esu'e gagana fa'anatura, o se seti o le CodeSearchNet Challenge ua saunia fa'aopoopo, e aofia ai
99 masani fesili e tusa ma le 4 faʻamatalaga faʻapitoa e faʻamatalaina ai le faʻaogaina o tulafono faʻapitoa i le CodeSearchNet Corpus dataset, e aofia ai le 6 miliona metotia ma galuega (seti tele e tusa ma le 20 GB). O le CodeSearchNet Challenge e mafai ona avea ma fa'ailoga mo le iloiloina o le aoga o nisi metotia mo le su'eina o le gagana fa'anatura. Fa'aaogā meafaigaluega KubeFlow saunia
faataitaiga code search engine.

puna: opennet.ru