Completenator: Advance Code Completion

Abstract

It is said that the number of programmers has doubled every five years since the 1980 s.[1] Especially after the pandemic, more and more people are getting into the IT sector considering the current evolution in technology and increased automation. Code Completion has always been a household name for programmers. It is a feature that allows the system to anticipate what you will type next. It has been around for a while, assisting the developers whenever they get stuck. It is quite evident that Code Completion no t only saves time but also increases the product ivity of the developer. Recently there are a lot of Auto Code Completion models that do code completion but in a limited way. They are only successful in predicting the next tokens for a shorter code sequence and falter when it comes to a larger code sequence. We have proposed our solution to overcome this crisis using Open AI's GPT-2 model. which is based on transformer architecture . It is a pre-trained model specifically used for traditional language modeling, i.e. predicting the next words in the token. We fine-tuned the model and re-trained it with python source code files cloned from a GitHub repository. Furthermore, we evaluated the model performance. The results obtained were better than the other tradition al methodologies. The goal of our system was to automatically complete the code where it not only predicts the next set of words but an entire sentence. And our system was able to do this on general-purpose python code.

Keywords: Code Completion, Natural Language Pr eprocessing, GPT-2, Language Models