GeneticSharp in the Wild: Context-Sensitive Code Completion
For this second post the chosen one is the
Completion: Improving Predictions with Genetic Algorithms.
The main motivation of the author about this paper was: Current methods of training code completion systems can possibly be improved in order to reduce prediction errors. This requires that the amount of information considered in a pattern is optimized. The question raised is then: how should the training of predictive models be focused in order to increase prediction quality?
This thesis will aim to answer the question: to what extent is it possible to improve predictions of existing state-of-the-art code completion systems with a genetic algorithm?
GeneticSharp is an open-source Genetic Algorithm library for C#, released under the MIT license (Giacomelli). It has an extensible interface that allows for most, if not all, functionality to be implemented from scratch via interfaces or leveraged by extending base classes. Classes and interfaces also use the same terminology that has already been established, which makes the translation from theory to implementation much more clear.
As a result of the new training scheme, the quality of predictions can be increased without losing generalizability. Application of the new training scheme could possibly be applied to any code completion systems that trains a predictive model, making it a candidate for improving existing systems as well as in future research.
results comparison between GCC and GeneCSCC (developed using GeneticSharp)
Marcus Ording wrote the paper for his degree project in Computer Engineering at KTH Royal Institute of Technology from Stockholm, Sweden.
You can access the full paper directly on DiVA Portal.