GitHub Actions: データサイエンティストに新たな超能力を提供

A Gentle Introduction To GitHub Actions What Are GitHub Actions?

Example: A fastpages Action Workflow Part 1: Define Workflow Triggers

Conclusion Related Materials Getting In Touch Footnotes

Hi, I’m Hamel Husain. I’m a machine learning engineer at GitHub. Recently, GitHub released a new product called GitHub Actions, which has mostly flown under the radar in the machine learning and data science community as just another continuous integration tool.

Recently, I’ve been able to use GitHub Actions to build some very unique tools for Data Scientists, which I want to share with you today. Most importantly, I hope to get you excited about GitHub Actions, and the promise it has for giving you new superpowers as a Data Scientist. Here are two projects I recently built with Actions that show off its potential:

fastpages is an automated, open-source blogging platform with enhanced support for Jupyter notebooks. You save your notebooks, markdown, or Word docs into a directory on GitHub, and they automatically become blog posts. Read the announcement below:

We're launching fastpages, a platform which allows you to host a blog for free, with no ads. You can blog with @ProjectJupyter notebooks, @office Word, directly from @github's markdown editor, etc.Nothing to install, & setup is automated!https://t.co/dNSA0oQUrN

Wouldn’t it be cool if you could invoke a chatbot natively on GitHub to test your machine learning models on the infrastructure of your choice (GPUs), log all the results, and give you a rich report back in a pull request so that everyone could see the results? You can with GitHub Actions!

Consider the below annotated screenshot of this Pull Request:

A more in-depth explanation about the above project can be viewed in this video:

Using GitHub Actions for machine learning workflows is starting to catch on. Julien Chaumond, CTO of Hugging Face, says:

GitHub Actions are great because they let us do CI on GPUs (as most of our users use the library on GPUs not on CPUs), on our own infra! 1

Additionally, you can host a GitHub Action for other people so others can use parts of your workflow without having to re-create your steps. I provide examples of this below.

A Gentle Introduction To GitHub Actions

GitHub Actions allow you to run arbitrary code in response to events. Events are activities that happen on GitHub such as:

When an event is created, the GitHub Actions context is hydrated with a payload containing metadata for that event. Below is an example of a payload that is received when an issue is created:

{ "action": "created", "issue": { "id": 444500041, "number": 1, "title": "Spelling error in the README file", "user": { "login": "Codertocat", "type": "User", }, "labels": [ { "id": 1362934389, "node_id": "MDU6TGFiZWwxMzYyOTM0Mzg5", "name": "bug", } ], "body": "It looks like you accidently spelled 'commit' with two 't's." } This functionality allows you to respond to various events on GitHub in an automated way. In addition to this payload, GitHub Actions also provide a plethora of variables and environment variables that afford easy to access metadata such as the username and the owner of the repo. Additionally, other people can package useful functionality into an Action that other people can inherit. For example, consider the below Action that helps you publish python packages to PyPi:

The Usage section describes how this Action can be used:

• name: Publish a Python distribution to PyPI uses: pypa/gh-action-pypi-publish@master with: user: __token__ password: ${{ secrets.pypi_password }} This Action expects two inputs: user and a password. You will notice that the password is referencing a variable called secrets, which is a variable that contains an encrypted secret that you can upload to your GitHub repository. There are thousands of Actions (that are free) for a wide variety of tasks that can be discovered on the GitHub Marketplace. The ability to inherit ready-made Actions in your workflow allows you to accomplish complex tasks without implementing all of the logic yourself. Some useful Actions for those getting started are:

actions/checkout: Allows you to quickly clone the contents of your repository into your environment, which you often want to do. This does a number of other things such as automatically mount your repository’s files into downstream Docker containers.

mxschmitt/action-tmate: Proivdes a way to debug Actions interactively. This uses port forwarding to give you a terminal in the browser that is connected to your Actions runner. Be careful not to expose sensitive information if you use this.

actions/github-script: Gives you a pre-authenticated ocotokit.js client that allows you to interact with the GitHub API to accomplish almost any task on GitHub automatically. Only these endpoints are supported (for example, the secrets endpoint is not in that list).

In addition to the aforementioned Actions, it is helpful to go peruse the official GitHub Actions docs before d