Avaliando as redes neurais recorrentes na complementação de código

Abstract

Code completion is a desirable feature for Integrated development environment. Several are the challenges encountered in its implementation. Languages present innumerable classes, methods, interfaces and resources. Therefore, recommending or indicating commands, instructions implies analyzing countless possibilities. The code completion functionality allows for better assertiveness during the development process. In terms of applied methods, the systematic literature review strongly indicated the use of natural language processing techniques, among which neural models showed a high rate of use. The recurrent neural network stand out for their ability to consider previous values when generating the prediction, this ability was strongly indicated as fruitful for code completion. This project aims to evaluate variations of recurrent neural network for completion tasks for the Python programming language focused on recommendations of application programming interface . For this purpose, several variations of recurrent neural network will be tested. [resumo fornecido pelo autor]