Source Code Embeddings

A set of six pretrained fastText models for semantic representations of source code. 

Each of the models has been trained on high-quality GitHub repositories where the primary language is one of Java, Python, C++, C#, C, PHP. For collecting training data 13.144 repositories were cloned, 2.402.790.348 lines of code were read out of 944,467,560 files and preprocessed, to finally produce a total of 944.467.560 tokens of clean training data. 

For further details refer to the following paper: 

Efstathiou, V.,  Spinellis, D., 2019. "Semantic Source Code Models Using Identifier Embeddings". In 16th International Conference on Mining Software Repositories: Data Showcase Track. MSR'19.