Glyph

Glyph uses Machine Learning and Natural Language Processing to understand commit messages. This knowledge can be used for classifying commits into categories such as Bug-fixes, Feature additions, Improvements etc.

• Using Glyph with Kebechet, smart CHANGELOG entries out of commit messages can be generated.
• Glyph can also be used as a standlone library for analyzing commits from a locally stored repository (see usage below)

Running this project from Git

git clone [email protected]:thoth-station/glyph.git  # or use https
cd glyph
pipenv install --dev
PYTHONPATH=. pipenv run ./thoth-glyph --help

Installing this project from PyPI

This project is available on PyPI, to install it:

pip install thoth-glyph

Features

• Commit Classification: Singular commits can be classified using the following command:

  thoth-glyph classify -m "COMMIT MESSAGE TO BE ANALYZED"

• Classifying Multiple Commits: Mulitple commit can be classified together using the classify-repo command. By default, this action classifies all the commits in the repository. Optionally, a date-range (YYYY-MM-DD) can be provided:

  thoth-glyph classify-repo --path /path/to/git/repo --start 2020-05-01 --end 2020-05-10

• Classifying Using Tags: Commits can also be picked using git tags. The following command will pick commits between the tags v3.7.1 and v3.7.2

  thoth-glyph classify-repo-by-tag --path /path/to/git/repo --start_tag v3.7.1 --end_tag v3.7.2

Sample Usage

$ thoth-glyph classify -m "Fixed server bug that impacted performance"
2020-08-12 19:45:47,798 4594 WARNING  thoth.common:346: Logging to a Sentry instance is turned off
2020-08-12 19:45:47,799 4594 INFO     thoth.common:368: Logging to rsyslog endpoint is turned off
2020-08-12 19:45:47,799 4594 INFO     glyph:68: Version: 0.0.0
2020-08-12 19:45:47,800 4594 INFO     glyph:83: Classifying commit
2020-08-12 19:45:47,800 4594 INFO     thoth.glyph.models:33: Model Path : /home/tussharm/.local/lib/python3.6/site-     packages/thoth/glyph/data/model_commits_v2_quant.bin
Label : corrective

$ thoth-glyph classify-repo --path /home/tussharm/fork/glyph/ --start 2020-08-08 --end 2020-08-12
2020-08-12 19:51:26,743 4873 WARNING  thoth.common:346: Logging to a Sentry instance is turned off
2020-08-12 19:51:26,743 4873 INFO     thoth.common:368: Logging to rsyslog endpoint is turned off
2020-08-12 19:51:26,744 4873 INFO     glyph:68: Version: 0.0.0
2020-08-12 19:51:26,744 4873 INFO     glyph:100: Classifying commits in the given date-range
2020-08-12 19:51:26,749 4873 INFO     thoth.glyph.models:44: Model Path : /home/tussharm/.local/lib/python3.6/site-p    packages/thoth/glyph/data/model_commits_v2_quant.bin
2020-08-12 19:51:26,768 4873 INFO     thoth.glyph.models:52: 6 commits classified
                                           message labels_predicted
0                                 readme updated #27       perfective
1  merge pull request #1 from tushar7sharma/commi...    nonfunctional
2  merge remote-tracking branch 'upstream/master'...         features
3  grouping user-defined commit phrases (#28)* co...         features
4  commits can be collected inside user-defined g...         features
5  merge remote-tracking branch 'upstream/master'...         features

Integration with Kebechet

Kebechet can use Glyph by reading the project's configuration from .thoth.yaml file. Glyph's supported formatters and ML classifers can be specified in this configuration file.

• See sample manager configuration here
• See sample changelog generated using Glyph here

Model and Dataset

Currently Glyph ships with a model trained using Facebook's fasttext library over a dataset of ~5000 commits collected from multiple large-scale open source projects (see referred publications for more details). The library can be easily extended to accomodate more models. Developers are welcome to contribute and improve the classification accuracy.

References

• https://arxiv.org/pdf/1711.05340.pdf
• http://labsoft.dcc.ufmg.br/lib/exe/fetch.php?media=cibse-geanderson.pdf
• https://github.com/gesteves91/fasttext-commit-classification
• https://github.com/nxs5899/Multi-Class-Text-Classification----Random-Forest