CoChangeAnalysis

Analyzes the git structure of a project to find the best configuration of co-change detection.

Also analyzes output from Market Basket Analysis, Dynamic Time Warping and CoCo.

Utilizes smells reported by https://github.com/darius-sas/astracker (ASTracker)

Related to https://github.com/RonaldKruizinga/CoSmellingChanges (CoCo)

Requirements

The requirements to use this script are:

• Python 3.7
• Pip

Installation

The source code for this project can be cloned using Git or downloaded from Github.

The required packages can be installed via the included Pipfile.

Usage

The analysis can be run in two ways. The first is by running Main.py. This file has a few different methods referenced for various purposes.

• hyper_param_analysis() and threshold_distribution() are used to generate hyperparameters for CoCo. threshold_distribution() requires CoCo to have run with a threshold of 1 first, so that it can calculate the distribution.

• generate_analysis() is used to run the market basket analysis and dynamic time warping

• run_exploration() runs a data exploration on all three algorithm, also using the output from ASTracker. For large projects, this should be done on a server cluster with a significant amount of memory. Example scripts for this can be found in the execution_scripts folder. These examples are used to run the project on the Peregrine Cluster of the University of Groningen.

• results_analysis() analyses the results and provides some interesting plots.

Configuration

In order to use the application to analyse hyperparameters for a project, certain properties in the config.py, located in the main directory, need to be set.

An example is left in the config for the Sonarlint-IntelliJ project.

• A github api key is required in order to recover the dates of analyzed commits
• A start and end date of the analysis are required as a time range.
• Project information, such as name, url, branch and owner are required
• Directories, such as input, output and cloning directories can be configured

Input

As input the application can require the output from CoCo and ASTracker.

Output

The system outputs the co-changes detected using the MBA and DTW algorithms, as well as statistics regarding the overlap of smells with co-changes.

Structure

• helper_scripts provide utility functions
• hyperparameter_analysis provides all necessary functions for that purpose
• model contains the classes defined
• results_analysis contains code for plotting interesting plots.