Transformer GPT-2 Model for Code Prediction in Machine Learning Projects

Pan T.P.W.; Sanchez L.M.C. · Communications in Computer and Information Science (2025) · DOI: 10.1007/978-3-031-91428-7_17

Resumen

Machine Learning projects have increased in their application due to the different demands for sophisticated solutions in companies and fields of study in recent years. This increase has led to a profound transformation in the nature of Machine Learning project development involving complex steps which require considerable dedication of resources. In this paper we present a model based on GPT-2 which will be implemented and trained to perform task-specific source code generation. Using a dataset of different tasks in a Machine Leaning project, we obtain the necessary data for training the models for each specific task class. Then, we use several evaluation metrics to verify the quality of the trained models. The results show that it is feasible to perform the task of automatic source code generation and improvements needed for the future project. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Instituciones: Universidad Peruana de Ciencias Aplicadas