Ensemble Learning for Method-Call Recommendation

One of the tool systems commonly used by developers is code completion - an assistant for developers to increase typing speed, explore new APIs, handle errors and just make life easier. Current research tries to improve on existing tools by exploiting implicit information and make it available to developers. This explores various context information that can be used to make code completion assistants better. We want to answer the question what context information is best suited for code completion, whether this is type-dependent, and how they can be combined in a joint model. In the first part, we introduce the subject and discuss the state of the art, then introduce terminology we use later in the book. In the second part, we present the context information that we use to predict a developer's intent. We then evaluate the models we get using this context information. In the third part, we focus on combining those models to further improve the proposals of the code completion.