Skip to content

Latest commit

 

History

History
198 lines (141 loc) · 9.75 KB

CONTRIBUTING.md

File metadata and controls

198 lines (141 loc) · 9.75 KB
Raw

Contribution Guidelines

It’s an honor to have you on board!

We are proud of this project and have been working to make it great since day one. We believe you will love it, and we know there’s room for improvement. We want to

  • implement features that make what you want to do possible and/or easy.
  • write more tutorials and examples that help you get familiar with Optuna.
  • make issues and pull requests on GitHub fruitful.
  • have more conversations and discussions on Gitter.

We need your help and everything about Optuna you have in your mind pushes this project forward. Join Us!

If you feel like giving a hand, here are some ways:

  • Implement a feature
    • If you have some cool idea, please open an issue first to discuss design to make your idea in a better shape.
  • Send a patch
  • Report a bug
    • If you find a bug, please report it! Your reports are important.
  • Fix/Improve documentation
    • Documentation gets outdated easily and can always be better, so feel free to fix and improve
  • Let us and the Optuna community know your ideas and thoughts.
    • Contribution to Optuna includes not only sending pull requests, but also writing down your comments on issues and pull requests by others, and joining conversations/discussions on Gitter.
    • Also, sharing how you enjoy Optuna is a huge contribution! If you write a blog, let us know about it!

If you write code, we have some conventions as follows.

Guidelines

Setup Optuna

First of all, fork Optuna on GitHub. You can learn about fork in the official documentation.

After forking, download and install Optuna on your computer.

git clone [email protected]:YOUR_NAME/optuna.git
cd optuna
pip install -e .

Checking the Format, Coding Style, and Type Hints

Code is formatted with black, and docstrings are formatted with blackdoc. Coding style is checked with flake8 and isort, and additional conventions are described in the Wiki. Type hints, PEP484, are checked with mypy.

You can check the format, coding style, and type hints at the same time just by executing a script formats.sh. If your environment is missing some dependencies such as black, blackdoc, flake8, isort or mypy, you will be asked to install them.

You can also check them using tox like below.

$ pip install tox
$ tox -e flake8 -e black -e blackdoc -e isort -e mypy

If you catch format errors, you can automatically fix them by auto-formatters.

# Install auto-formatters.
$ pip install ".[checking]"

$ ./formats.sh

Documentation

When adding a new feature to the framework, you also need to document it in the reference. The documentation source is stored under the docs directory and written in reStructuredText format.

To build the documentation, you need to run:

pip install -e ".[document]"

Note that the above command might try to install PyTorch without CUDA to your environment even if your environment has CUDA version already.

Then you can build the documentation in HTML format locally:

cd docs
make html

HTML files are generated under build/html directory. Open index.html with the browser and see if it is rendered as expected.

Optuna's tutorial is built with Sphinx-Gallery and some other requirements like LightGBM and PyTorch meaning that all .py files in tutorial directory are run during the documentation build if there's no build cache. Whether you edit any tutorial or not doesn't matter.

To avoid having to run the tutorials, you may download executed tutorial artifacts named "tutorial" from our CI (see the capture below) and put them in docs/build before extracting the files in the zip to docs/source/tutorial directory.

image

Writing a Tutorial Tutorials are part of Optuna’s documentation. Optuna depends on Sphinx to build the documentation HTML files from the corresponding reStructuredText (.rst) files in the docs/source directory, but as you may notice, Tutorial directory does not have any .rst files. Instead, it has a bunch of Python (.py) files. We have Sphinx Gallery that executes those .py files and generates .rst files with standard outputs from them and corresponding Jupyter Notebook (.ipynb) files. These generated .rst and .ipynb files are written to the docs/source/tutorial directory. The output directory (docs/source/tutorial) and source (tutorial) directory are configured in sphinx_gallery_conf of docs/source/conf.py. These generated .rst files are handled by Sphinx like the other .rst files. The generated .ipynb files are hosted on Optuna’s documentation page and downloadable (check Optuna tutorial).

The order of contents on tutorial top page is determined by two keys: one is the subdirectory name of tutorial and the other is the filename (note that there are some alternatives as documented in Sphinx Gallery - sorting, but we chose this key in https://github.com/optuna/optuna/blob/2e14273cab87f13edeb9d804a43bd63c44703cb5/docs/source/conf.py#L196). Optuna’s tutorial directory has two directories: (1) 10_key_features, which is meant to be aligned with and explain the key features listed on README.md and (2) 20_recipes, whose contents showcase how to use Optuna features conveniently. When adding new content to the Optuna tutorials, place it in 20_recipes and its file name should conform to the other names, for example, 777_cool_feature.py. In general, please number the prefix for your file consecutively with the last number. However, this is not mandatory and if you think your content deserves the smaller number (the order of recipes does not have a specific meaning, but in general, order could convey the priority order to readers), feel free to propose the renumbering in your PR.

You may want to refer to the Sphinx Gallery for the syntax of .py files processed by Sphinx Gallery. Two specific conventions and limitations for Optuna tutorials:

  1. 99 #s for block separation as in https://github.com/optuna/optuna/blob/2e14273cab87f13edeb9d804a43bd63c44703cb5/tutorial/10_key_features/001_first.py#L19
  2. Execution time of the new content needs to be less than three minutes. This limitation derives from Read The Docs. If your content runs some hyperparameter optimization, set the timeout to 180 or less. You can check this limitation on Read the Docs - Build Process.

Unit Tests

When adding a new feature or fixing a bug, you also need to write sufficient test code. We use pytest as the testing framework and unit tests are stored under the tests directory.

You can run all your tests as follows:

# Run all the unit tests.
pytest

# Run all the unit tests defined in the specified test file.
pytest tests/${TARGET_TEST_FILE_NAME}

Continuous Integration and Local Verification

Optuna repository uses GitHub Actions and CircleCI.

Currently, we are migrating to GitHub Actions but still we use CircleCI for testing document because it makes it much easier to check built documentation.

Local Verification

By installing act and Docker, you can run tests written for GitHub Actions locally.

JOB_NAME=checks
act -j $JOB_NAME

Currently, you can run the following jobs: documentation and doctest may not be executable depending on your choice of docker image of act.

  • checks
    • Checking the format, coding style, and type hints
  • tests
    • Runs unit tests
  • documentation
    • Builds documentation including tutorial
  • doctest
    • Runs doctest

Creating a Pull Request

When you are ready to create a pull request, please try to keep the following in mind:

Title

The title of your pull request should

  • briefly describe and reflect the changes
  • wrap any code with backticks
  • not end with a period

The title will be directly visible in the release notes.

Example

Introduces Tree-structured Parzen Estimator to optuna.samplers

Description

The description of your pull request should

  • describe the motivation
  • describe the changes
  • if still work-in-progress, describe remaining tasks