Visualizing risky code: bringing deep learning to your IDE


Source code often contains bugs, all programmers and users know it. In this talk we present an alternative approach to find them! Btw, how do you find “bugs” in language? If, e.g., one says “Hi, how is you?” Well, even if you don’t know the grammar behind it, it sounds strange.
Most of the people say “Hi, how are you?”, not “Hi, how is you?”. We want to copy this idea to find strange code in programming languages.
We created a model of Java code, based on thousands of projects present on GitHub. This model can be seen as a summary of all the code we found. If we now feed our model with a piece of code, it can tell us how often it already has seen such code and it can give us a measure of how strange this code is , since it was rarely or never written by anybody else.
In this talk we present a Visual Studio Code plugin, that visualizes the strangeness of the code currently written: it warns the programmer about code that nobody before has written before it in that way. Of course, this can be because the source code is a masterpiece, but it might also be that the code is buggy. We hope that this can be of help to find and fix bugs as they arise.
The plugin is currently available on !