LSTM Based Source Code Generation for Turkish Pseudo Code of Algorithm

Abstract

Algorithmic thinking and programming abilities of students is controversial and popular issue in technological education programs in schools and universities. Students that have not best mathematical and analytical background may have difficulties in learning computer programing. Moreover, learning programming is highly difficult for a single individual to establish connection between discrete pseudo code of algorithm and source code. Another problem is required time to write a piece of program code. In order to solve this problem, there are some tools that tutor students to get analyze and realize relation between pseudo code and source code. In this study, we propose a deep learning method that is Long Short Term-Memory (LSTM) based source code generator from Turkish pseudo codes. For this purpose, we used Introduction to programming course exams in vocational high school as dataset to train LSTM. When users query a Turkish pseudo code of algorithm, C# source code is generated. In order to measure success of proposed system, generated source code and instructor’s source code is analyzed with text similarity methods. Results show that proposed system is useful for students to learn fundamental programming skills.

Keywords

• otomatik kod üretimi
• derin öğrenme
• uzun kısa süreli bellek
• automatic code generation
• deep learning
• long short term-memory