Understanding your code with Machine Learning

The Data Platform for your �Software Development Life Cycle

Intro

​

Hugo, Data Scientist @ source{d}

​

Alex, Engineer @ source{d}

2

Agenda

• Goals of this workshop
• MLonCode field overview
• Data Science project
• Workshop tasks
• Hands On: preparations�

3

Goals of the workshop

What to expect from the rest 2h:

• Example of running a data-science project
• Specifics of Machine Learning on Code field
• OSS tooling to process the source code
• Dataset and all needed software is on USB*

​

*Start copying the data!

4

Machine Learning on Code

Overview

5

Data Science project

Main activities during a data science project

1. Problem statement
2. Data collection
3. Exploration
4. Evaluation
5. Communication

6

Workshop tasks

​

• Developer & project similarity

​

• Function name suggestion

7

Workshop tasks

Developer and project similarity

• How to find similar developers, based on their contributions?

​

• How to find similar projects, base on their topics?

​

Both can be done in unsupervised way,

by building a vector representation for individual developer & project.

8

Workshop tasks

Function name suggestion

How to predict the name of a function, based on its body text?

We are going to use a simple Machine Translation baseline* model

​

9

* check github.com/src-d/awesome-machine-learning-on-source-code/ for State Of the Art (SOtA) models

Hands On: preparations

Before we begin: necessary steps to prepare local environment

• After you have copied the data
• Check REAMDE.md for instructions on how to proceed
• Load the 3 required Docker images
• Open first Notebook

10

Let’s get coding!