DeepAPI#: CLR/C# call sequence synthesis from text query

Alexander Chebykin, Mikhail Kita, Iakov Kirilenko

Research output

Abstract

Developers often search for an implementation of typical features via libraries (for example, how to create a UI button control, extract data from a JSON-formatted file, etc.). The Internet is the usual source of the information on the topic. However, various statistical tools provide an alternative: after processing large amounts of source code and learning common patterns, they can convert a user request to a set of relevant function calls. We examine one of those tools - DeepAPI. This fresh deep learning based algorithm outperforms all others (according to its authors). We attempt to reproduce this result using different target programming language - C# - instead of Java used in the original DeepAPI. In this paper we report arising problems in the data gathering for training, difficulties in the model construction and training, and finally discuss possible modifications of the algorithm.

Original languageEnglish
JournalCEUR Workshop Proceedings
Volume1864
Publication statusPublished - 1 Jan 2017
Event2nd Conference on Software Engineering and Information Management, SEIM 2017 - Saint Petersburg
Duration: 21 Apr 2017 → …

Fingerprint

Computer programming languages
Internet
Processing
Deep learning

Scopus subject areas

  • Computer Science(all)

Cite this

@article{e29acdd4fe8e45c1b47c648059a18b75,
title = "DeepAPI#: CLR/C# call sequence synthesis from text query",
abstract = "Developers often search for an implementation of typical features via libraries (for example, how to create a UI button control, extract data from a JSON-formatted file, etc.). The Internet is the usual source of the information on the topic. However, various statistical tools provide an alternative: after processing large amounts of source code and learning common patterns, they can convert a user request to a set of relevant function calls. We examine one of those tools - DeepAPI. This fresh deep learning based algorithm outperforms all others (according to its authors). We attempt to reproduce this result using different target programming language - C# - instead of Java used in the original DeepAPI. In this paper we report arising problems in the data gathering for training, difficulties in the model construction and training, and finally discuss possible modifications of the algorithm.",
author = "Alexander Chebykin and Mikhail Kita and Iakov Kirilenko",
year = "2017",
month = "1",
day = "1",
language = "English",
volume = "1864",
journal = "CEUR Workshop Proceedings",
issn = "1613-0073",
publisher = "RWTH Aahen University",

}

DeepAPI# : CLR/C# call sequence synthesis from text query. / Chebykin, Alexander; Kita, Mikhail; Kirilenko, Iakov.

In: CEUR Workshop Proceedings, Vol. 1864, 01.01.2017.

Research output

TY - JOUR

T1 - DeepAPI#

T2 - CLR/C# call sequence synthesis from text query

AU - Chebykin, Alexander

AU - Kita, Mikhail

AU - Kirilenko, Iakov

PY - 2017/1/1

Y1 - 2017/1/1

N2 - Developers often search for an implementation of typical features via libraries (for example, how to create a UI button control, extract data from a JSON-formatted file, etc.). The Internet is the usual source of the information on the topic. However, various statistical tools provide an alternative: after processing large amounts of source code and learning common patterns, they can convert a user request to a set of relevant function calls. We examine one of those tools - DeepAPI. This fresh deep learning based algorithm outperforms all others (according to its authors). We attempt to reproduce this result using different target programming language - C# - instead of Java used in the original DeepAPI. In this paper we report arising problems in the data gathering for training, difficulties in the model construction and training, and finally discuss possible modifications of the algorithm.

AB - Developers often search for an implementation of typical features via libraries (for example, how to create a UI button control, extract data from a JSON-formatted file, etc.). The Internet is the usual source of the information on the topic. However, various statistical tools provide an alternative: after processing large amounts of source code and learning common patterns, they can convert a user request to a set of relevant function calls. We examine one of those tools - DeepAPI. This fresh deep learning based algorithm outperforms all others (according to its authors). We attempt to reproduce this result using different target programming language - C# - instead of Java used in the original DeepAPI. In this paper we report arising problems in the data gathering for training, difficulties in the model construction and training, and finally discuss possible modifications of the algorithm.

UR - http://www.scopus.com/inward/record.url?scp=85025164732&partnerID=8YFLogxK

M3 - Conference article

AN - SCOPUS:85025164732

VL - 1864

JO - CEUR Workshop Proceedings

JF - CEUR Workshop Proceedings

SN - 1613-0073

ER -