This repository contains the GAP tool presented in the companion research article GAP: Forecasting Commit Activity in git Projects published in 2019 in the Journal on Systems and Software. GAP analyses and forecasts developer activity in the form of git commits by relying on a simple probabilistic model.
The tool has been developed by Alexandre Decan, principal researcher at the Software Engineering Lab of the University of Mons (Belgium) in the context of the SECO-ASSIST Excellence of Science research project.
Abandonment of active developers poses a significant risk for many open source software projects. This risk can be reduced by forecasting the future activity of contributors involved in such projects. Focusing on the commit activity of individuals involved in git repositories, we propose a practical probabilistic forecasting model that has been validated on a wide variety of projects accounting for 7,528 git repositories and 5,947 active contributors. This model is implemented as part of the GAP open source tool, that predicts future commit activity.
The tool is available at the root of this repository (gap.py). The tool can be downloaded and used as a Python script (the usual way) or installed through pip (pip install git+https://github.com/AlexandreDecan/gap). The dependencies will be automatically installed.
GAP is released under LGPLv3 - GNU Lesser General Public License, version3.
$ gap -h
usage: gap [-h] [--date DATE] [--obs OBS]
[--probs [PROB [PROB ...]]]
[--limit LIMIT] [--mapping [MAPPING]]
[--branches [BRANCH [BRANCH ...]]]
[--as-dates]
[--text | --csv | --json | --plot [PLOT]]
[PATH [PATH ...]]
In its simplest form, GAP only requires a path to the git repositories of the software projects that need to be analysed. One or more such repositories can be provided as input, depending on the scope of the analysis. For example, a company may desire to analyse the commit activity of all its paid contributors in the open source projects the company is involved in; a package maintainer may desire to analyse the global activity of all the packages he is maintaining; a package developer may wish to analyse the activity in a project and all its dependents; and so on.
By default, GAP analyses all branches of each repository but the list of branches to analyse can be specified. Optionally, a list mapping multiple authors into specific identities can be provided. Such mapping files can be used to support identity merging, data anonymization, or to exclude specific authors from the analysis. Parameters can be provided to set the date of the analysis (current date by default), the number of observations used by the model (20 by default), the list of probability values (0.5, 0.6, 0.7, 0.8 and 0.9 by default), and the minimal recency of the last activity (30 days by default). Predictions can be expressed either as dates or as relative time differences (by default).
$ gap ./repo --date 2019-05-16 --mapping anon.csv
last 0.5 0.6 0.7 0.8 0.9
author
D. Young 0 1 1 2 3 4
Z. Andrews 0 4 5 8 10 23
D. Johnson 0 2 3 4 4 8
J. Berry 0 3 6 7 29 131
B. Rodriguez 0 1 1 3 3 6
bots (grouped) 0 1 1 1 1 2
M. Johnston 0 3 3 4 6 7
L. Owen -1 3 7 7 17 22
S. Allen -2 3 7 12 25 38
J. Schultz -4 5 10 13 23 38
J. Smith -4 0 3 6 7 21
M. Fry -9 -6 -3 0 8 21
J. Lopez -17 -11 -11 -9 -6 -1
S. Lewis -20 -12 -10 -3 50 96
The output shows, for each recently active contributor (first column), the time difference in days since the last known commit activity (second column), and the expected number of days until the next predicted day of activity according to a certain probability threshold ranging between
To ease their reuse by other automated tools, the forecasts can be exported into four different formats: (i) simple text, (ii) comma-separated values (csv), (iii) JSON, and (iv) bar activity charts. GAP output results can also be used as a basis for a project-level dashboard that visualises the project's past and estimated future commit activities.