This project is about establishing a complete profile of developers based on their open-source contributions.

Creating a service to allow people, typically recruiters from companies or some open-source projects, to search for a developer with a specific set of skills and that the service outputs a list of developers ranked by best match with a detailed profile accessible for each of them is the main project goal.

This project, proposed by Robin Hahling, started as a project for a Big Data class given at EPFL for the master of science in computer science.

The main goal is to build a database of developers, along with an evaluation of their programming skills, and make it available through a freely accessible API. This database could then be used by recruiters for companies or members of open source projects to find developers with the specific skills set they are looking for. The devmine-web project is a web front-end that accesses this API.

Data used is first collected from GitHub but other data source will be considered later on (BitBucket, SourceForce, Gitorious, ...) and maybe even data collected from other kind of data source such as user reputation from StackOverflow, academic background from LinkedIn and so on.

The main idea behind this project is that someone's contributions to open-source projects can be a valuable source of information for recruiters. A website that could highlight the skills of a developer, in a relatively good manner, by not taking into consideration only his academic formation or past employers but also actual code contributions and interactions within open-source communities would probably reveal being valuable source of information.

The project is divided into two main parts: the offline and the online part. The offline part is responsible of computing features and it is implemented in the devmine-features repository. The online part provides the API and lays in the devmine module from this repository.

On the offline part, each developers gets assigned a score for each feature, which value is normalized between 0 and 1. There are two different kind of features: the primitive features and the derived features. Primitive features can be used straight from the data source or require minimal processing, for instance: lines of code in language X, number of GitHub followers, whether the developer is available for hiring and so on. Derived features typically require some calculation. For instance, reputation of the developer determined by the PageRank algorithm run with followers/following developers from GitHub as links or HITS with developers as authorities and projects as hubs and so on. Each feature defines a function to compute it and a function to normalize it and is highly dependent on the available data sources.

On the online part, queries can be split into two parts: ranking and filtering. The ranking part can be viewed as the definition of the "perfect developer", or an assignment of weight to certain features. More precisely, a search query consists of a set a feature names with associated weights provided by the user that sends the query through a frontend (typically the web frontend). Ranking is computed as a dot product between the scores matrix and the weight vector. The scores matrix is a matrix, preloaded into memory, which contains the developers as row and the features as column so each matrix cell correspond to the score of a developer for a particular feature. The weight vector is prefilled with default values for each feature but is modified according to the user query prior the dot production computation, which determines the ranking. Therefore, every single feature is taken into account to determine the ranking but only there weight differ. Filtering is used to specify attributes that developers must or must not have such as living in Switzerland, being fluent in Java or available for hiring.

The composition function, ie the dot product between the scores matrix and the weight vector, is used for the ranking. Once the ranking is established, results are filtered out by the filtering function.

This project uses mainly python in version 3. The easiest way to have every required libraries is to use a virtual environment:

• Install virtualenv if necessary.
• Set up a virtual environment: virtualenv -p python env (replace python with python3 if python 3 is not your default python version).
• Activate it: source env/bin/activate.
• Install the required libraries using pip: pip install -r requirements.txt
• When contributing to the project, you also need to install development requirements: pip install -r requirements_dev.txt
• Install a server backend. This should be corresponding to what you configure in the settings. Default is tornado: pip install tornado
• For a basic setup, run the following command: invoke setup
• If you do not need to tweak anything, simply run the application: python run.py

When contributing, make sure that your changes are conform to PEP8 by running invoke pep8. You may also want to do a static analysis of the code: invoke pyflakes. To run a full check (both PEP8 and static analysis), run: invoke check