Improving Code Autocompletion with Transfer Learning

Abstract

Software language models have achieved promising results predicting code completion usages, and several industry studies have described successful IDE integration. Recently, accuracy in autocompletion prediction improved 12.8%[2] from training on a real-world dataset collected from programmers’ IDE activities. But what if the number of examples of IDE autocompletion in the target programming language is inadequate for model training? In this paper, we highlight practical reasons for this inadequacy, and make a call to action in using transfer learning to overcome the issue.