Paper Reading Are Deep Neural Networks the Best Choice for Modeling Source Code
Summary: Carefully adapting n-gram models can yield better performance in code recommendation tasks than deep learning models.
Evaluation: Quite careful & solid work. Good summary of interesting despite obvious observation. Made meaningful and executable link between two domains. Giga-token corpus.
Takeaways: github.com/SLP-Team/SLP-Core MRR is a one-number metric for evaluation prediction results. (Mean Reciprocal Ranking) Yielding better results when combined with DL. extrinsic eval means using end results. intrinsic means using statistical measurements, like entropy.
What you can learn from this to make your research better?
Evaluation approach. What is a good sub-problem to attack.
Details and Problems From the presenters’ point of view, what questions might audience ask?