Deep Learning Based Code Completion Models for Programming Codes

With the fast development of Information Technology, program software and mobile applications have been widely used in the world, and are playing important roles in human's daily life. Thus, writing programming codes has been important work in many fields. however, it is a hard and time-cost task which presents a great amount of workload to programmers. To make programmers' work easier, intelligent code completion models have been a popular research topic in recent years. This paper designs Deep Learning based models to automatically complete programming codes, which are LSTM-based neural networks, and are combined with several techniques such as Word Embedding models in NLP (Natural Language Processing), and Multihead Attention Mechanism. Moreover, in the models, this paper raises a new algorithm of generating input sequences from partial AST (Abstract Syntax Tree) that have most relevance with nodes to be predicted which is named as RZT (Reverse Zig-zag Traverse) Algorithm, and is the first work of applying Multihead Attention Block into this task. This paper makes insight into codes of several different programming languages, and the models this paper presents show good performances in accuracy comparing with the state-of-art models.