In EDS ansehen

Leveraging deep learning for Python version identification

Gespeichert in:
Bibliographische Detailangaben
Titel: Leveraging deep learning for Python version identification
Autoren: Gerhold, Marcus, Solovyeva, Lola, Zaytsev, Vadim
Quelle: CEUR workshop proceedings. 3567:33-40
Verlagsinformationen: Rheinisch Westfälische Technische Hochschule, 2023.
Publikationsjahr: 2023
Schlagwörter: Deep Learning, CodeBERT, version identification, Python
Beschreibung: Python, recognized for its dynamic and adaptable nature, has found widespread application in a myriad of projects. As the language evolves, determining the Python version employed in a project becomes pivotal to ensure compatibility and facilitate maintenance. Deep learning (DL) has emerged as a promising tool to automate this process. In this research, we assess various DL techniques in determining the minimum Python version for a code snippet. We explore the complexities of handling Python data and the DL techniques to achieve high classification accuracy. Our experimental results show, that LSTM with CodeBERT embedding achives an accuracy of 92%. This success can be attributed to the LSTM's proficiency in capturing structural details of the hierarchical nature of a source code, complemented by CodeBERT's ability to discern contextual differences between keywords and variable names. This research provides insights into the challenges associated with utilizing programming languages for deep learning models and suggests potential solutions for addressing these issues. The envisioned applications extend to predicting the minimum required version for individual files or an entire code base.
Publikationsart: Article
Sprache: English
ISSN: 1613-0073
Zugangs-URL: https://research.utwente.nl/en/publications/f0f8b2f7-dc79-4515-a968-d55e36f0bbae
Dokumentencode: edsair.dris...02403..749cdc11b2fc3a4b2562daf39d0839b2
Datenbank: OpenAIRE
Schreiben Sie den ersten Kommentar!
Bitte loggen Sie sich zuerst ein