Skip to content

Latest commit

 

History

History
380 lines (272 loc) · 14.3 KB

CONTRIBUTING.md

File metadata and controls

380 lines (272 loc) · 14.3 KB

Contributing to Ignite

This project is a community effort, and everyone is welcome to contribute !

If you are interested in contributing to Ignite, there are many ways to help out. Your contributions may fall into the following categories:

  1. It helps us very much if you could

    • Report issues you’re facing
    • Give a 👍 on issues that others reported and that are relevant to you
    • Spread a word about the project or simply ⭐ to say "I use it"
  2. Answering queries on the issue tracker, investigating bugs and reviewing other developers’ pull requests are very valuable contributions that decrease the burden on the project maintainers.

  3. You would like to improve the documentation. This is no less important than improving the library itself! If you find a typo in the documentation, do not hesitate to submit a GitHub pull request.

  4. You would like propose a new feature and implement it

    • Post about your intended feature, and we shall discuss the design and implementation. Once we agree that the plan looks good, go ahead and implement it.
  5. You would like implement a feature or bug-fix for an outstanding issue

    • Look at the issues labelled as "help wanted"
    • Pick an issue and comment on the task that you want to work on this feature.
    • If you need more context on a particular issue, please ask and we shall provide.

Table of Contents

Developing Ignite

Quickstart guide for first-time contributors

  • Install miniconda for your system.
  • Create an isolated conda environment for pytorch-ignite:
conda create -n pytorch-ignite-dev python=3.8
  • Activate the newly created environment:
conda activate pytorch-ignite-dev
  • When developing please take care of preserving .gitignore file and make use of .git/info/exclude to exclude custom files like: .idea, .vscode etc.
  • Please refer to github first contributions guidelines and don't hesitate to ask the pytorch-ignite community in case of any doubt.
  • A good way to start is to tackle one of the good first issues.

Installation

  1. Make a fork of the repository on the GitHub (see here for details). As a result, for example your username is happy-ignite-developer, then you should be able to see your fork on the GitHub, e.g https://github.com/happy-ignite-developer/ignite.git

  2. Clone your fork locally and setup upstream. Assuming your username is happy-ignite-developer:

git clone https://github.com/happy-ignite-developer/ignite.git
cd ignite
git remote add upstream https://github.com/pytorch/ignite.git
git remote -v

You might see the following output:

origin  https://github.com/happy-ignite-developer/ignite.git (fetch)
origin  https://github.com/happy-ignite-developer/ignite.git (push)
upstream        https://github.com/pytorch/ignite (fetch)
upstream        https://github.com/pytorch/ignite (push)
  1. Sync and install all necessary dependencies:
git pull upstream master
python setup.py develop
pip install -r requirements-dev.txt
bash ./tests/run_code_style.sh install

Code development

Codebase structure

  • ignite - Core library files
    • engine - Module containing core classes like Engine, Events, State.
    • handlers - Module containing out-of-the-box handlers
    • metrics - Module containing out-of-the-box metrics
    • contrib - Contrib module with other metrics, handlers classes that may require additional dependencies
    • distributed - Module with helpers for distributed computations
  • tests - Python unit tests
  • examples - Examples and notebook tutorials
  • docs - Documentation files

If you modify the code, you will most probably also need to code some tests to ensure the correct behaviour. We are using pytest to write our tests:

New code should be compatible with Python 3.X versions. Once you finish implementing a feature or bugfix and tests, please run lint checking and tests:

Formatting Code

To ensure the codebase complies with a style guide, we use flake8 and ufmt (black and usort) to format and check codebase for compliance with PEP8.

Formatting without pre-commit

If you choose not to use pre-commit, you can take advantage of IDE extensions configured to black format or invoke black manually to format files and commit them.

To install flake8, ufmt and mypy, please run

bash ./tests/run_code_style.sh install

To format files and commit changes:

# This should autoformat the files
bash ./tests/run_code_style.sh fmt
# If everything is OK, then commit
git add .
git commit -m "Added awesome feature"
Formatting with pre-commit

To automate the process, we have configured the repo with pre-commit hooks to use µfmt to autoformat the staged files to ensure every commit complies with a style guide. This requires some setup, which is described below:

  1. Install pre-commit in your python environment.
  2. Run pre-commit install that configures a virtual environment to invoke ufmt and flake8 on commits.
pip install pre-commit
pre-commit install
  1. When files are committed:
    • If the stages files are not compliant with black or µsort, µfmt will autoformat the staged files. If this were to happen, files should be staged and committed again. See example code below.
    • If the staged files are not compliant with flake8, errors will be raised. These errors should be fixed and the files should be committed again. See example code below.
git add .
git commit -m "Added awesome feature"
# DONT'T WORRY IF ERRORS ARE RAISED.
# YOUR CODE IS NOT COMPLIANT WITH flake8, µsort or black
# Fix any flake8 errors by following their suggestions
# µfmt will automatically format the files so they might look different, but you'll need to stage the files
# again for committing
# After fixing any flake8 errors
git add .
git commit -m "Added feature"

Run tests:

To run a specific test, for example test_terminate from test_engine.py:

pytest tests/ignite/engine/test_engine.py -vvv -k test_terminate

To run all tests with coverage (assuming installed pytest-cov and pytest-xdist):

bash tests/run_cpu_tests.sh

On Windows, distributed tests should be skipped

SKIP_DISTRIB_TESTS=1 bash tests/run_cpu_tests.sh
Run distributed tests only on CPU

To run distributed tests only (assuming installed pytest-xdist):

export WORLD_SIZE=2
CUDA_VISIBLE_DEVICES="" pytest --dist=each --tx $WORLD_SIZE*popen//python=python tests/ -m distributed -vvv

Run Mypy checks:

To run mypy to check the optional static type:

bash ./tests/run_code_style.sh mypy

To change any config for specif folder, please see the file mypy.ini

Send a PR

If everything is OK, please send a Pull Request to https://github.com/pytorch/ignite from your fork.

If you are not familiar with creating a Pull Request, here are some guides:

NOTE : When sending a PR, please kindly check if the changes are required to run in the CI.

For example, typo changes in CONTRIBUTING.md, README.md are not required to run in the CI. So, please add [skip ci] in the PR title to save the resources. Ignite has setup several CIs.

  • GitHub Actions
  • Netlify

So, please add

  • [skip actions] for the changes which are not required to run on GitHub Actions,
  • [skip netlify] for the changes which are not required to run on Netlify PR Preview build, or
  • [skip ci] for the changes which are not required to run on any CI.

NOTE : Those skip statements are case sensitive, need open bracket [ and close bracket ]. And, Ignite has followed a convention of starting with skip word.

Sync up with the upstream

First, make sure you have set upstream by running:

git remote add upstream https://github.com/pytorch/ignite

Then you can see if you have set up multiple remote correctly by running git remote -v:

origin  https://github.com/{YOUR_USERNAME}/ignite.git (fetch)
origin  https://github.com/{YOUR_USERNAME}/ignite.git (push)
upstream        https://github.com/pytorch/ignite (fetch)
upstream        https://github.com/pytorch/ignite (push)

Now you can get the latest development into your forked repository with this:

git fetch upstream
git checkout master
git merge upstream/master

Writing documentation

Ignite uses Google style for formatting docstrings, specially from an example of Google style with Python 3 type annotations and

Examples: versionadded usage link, versionchanged usage link

Length of line inside docstrings block must be limited to 120 characters.

Local documentation building and deploying

Please, follow the instructions to build and deploy the documentation locally.

Install requirements
cd docs
pip install -r requirements.txt

Katex is also needed to build the documentation. To install katex, you need to have nodejs installed. Optionaly, we can install nodejs/npm using conda: conda install nodejs. Then you can install katex with npm or yarn (if installed).

npm install -g katex
# or if you use yarn package manager
yarn global add katex
Build
cd docs
make html
Local deployment

Please, use python 3.X for the command below:

cd docs/build
python -m http.server <port>
# python -m http.server 1234

Then open the browser at localhost:<port> (e.g. localhost:1234) and click to html folder.

Examples testing (doctests)

PyTorch-Ignite uses Sphinx directives. Every code that needs to be tested should be under .. testcode:: and expected output should be under .. testoutput::. For example:

.. testcode::

    def process_function(engine, batch):
        y_pred, y = batch
        return y_pred, y
    engine = Engine(process_function)
    metric = SSIM(data_range=1.0)
    metric.attach(engine, 'ssim')
    preds = torch.rand([4, 3, 16, 16])
    target = preds * 0.75
    state = engine.run([[preds, target]])
    print(state.metrics['ssim'])

.. testoutput::

    0.9218971...

If the floating point results are needed for assertion and the results can vary per operating systems and PyTorch versions, we could assert the results up to 4 or 6 decimal places and match the rest of the results with .... Learn more about sphinx.ext.doctest in the official documentation.

To make writing doctests easy, there are some configuratons defined in conf.py. Search doctest_global_setup in conf.py to see which variables and functions are available.

To run doctests locally:

cd docs
make html && make doctest