Deep Code Comment Generation, Xing Hu, Ge Li, Xin Xia, David Lo, Zhi Jin. Using a corpus of ~10k open source Java projects, this team created a neural network that can summarize methods and write human parity comments describing them. The team developed a novel AST vectorization for training by representing the tree as a sequence. Interestingly, this encoding phase is wholly reversible making this representation an ideal jumping off point for team's interested in training ML models off of source code.

results matching ""

    No results matching ""