possible completions. This demo presents a language agnostic machine learning
based Content Assist implementation.
Learning based Content Assist works by creating a model of the language by
training the Content Assist mechanism on a corpus of training source code and
then using the model to make predictions as the user is entering source code in
Objective evaluation metrics like key strokes per character achievable with this
approach as well result of subjective evaluation on real world code is shown.
The demo points out the additional possibilities opened up by using the learning
based approach including completion inside string literals and comments,
suggestion of variable names during definition and highly accurate multiple word
completions among others.
The approach presented uses a prediction model implementation based on variable
order markov modeling specifically the popular Prediction by Partial Match
method (PPM-C) algorithm.
The full text of the paper with details can be found here.
Rahul is software developer with ThoughtWorks Inc and lives in Providence, Rhode Island.