Timezone: »

Learning and Evaluating Contextual Embedding of Source Code
Aditya Kanade · Petros Maniatis · Gogul Balakrishnan · Kensen Shi

Wed Jul 15 08:00 AM -- 08:45 AM & Wed Jul 15 07:00 PM -- 07:45 PM (PDT) @ Virtual #None

Recent research has achieved impressive results on understanding and improving source code by building up on machine-learning techniques developed for natural languages. A significant advancement in natural-language understanding has come with the development of pre-trained contextual embeddings, such as BERT, which can be fine-tuned for downstream tasks with less labeled data and training budget, while achieving better accuracies. However, there is no attempt yet to obtain a high-quality contextual embedding of source code, and to evaluate it on multiple program-understanding tasks simultaneously; that is the gap that this paper aims to mitigate. Specifically, first, we curate a massive, deduplicated corpus of 6M Python files from GitHub, which we use to pre-train CuBERT, an open-sourced code-understanding BERT model; and, second, we create an open-sourced benchmark that comprises five classification tasks and one program-repair task, akin to code-understanding tasks proposed in the literature before. We fine-tune CuBERT on our benchmark tasks, and compare the resulting models to different variants of Word2Vec token embeddings, BiLSTM and Transformer models, as well as published state-of-the-art models, showing that CuBERT outperforms them all, even with shorter training, and with fewer labeled examples. Future work on source-code embedding can benefit from reusing our benchmark, and comparing against CuBERT models as a strong baseline.

Author Information

Aditya Kanade (Indian Institute of Science and Google Brain)
Petros Maniatis (Google Research)
Gogul Balakrishnan (Google)
Kensen Shi (Google)

More from the Same Authors