Continuous Machine Learning (CML) is an open-source library for implementing continuous integration & delivery (CI/CD) in machine learning projects. Use it to automate parts of your development workflow, including machine provisioning; model training and evaluation; comparing ML experiments across your project history, and monitoring changing datasets.

The iterative/setup-cml can be used as a GitHub Action to provide CML functions in your workflow. The action allows users to install CML without using the CML Docker container.

This action gives you:

• Access to all CML functions. For example:
  • cml comment create for publishing data visualization and metrics from your CI workflow as comments in a pull request.
  • cml pr create to open a pull request.
  • cml runner launch, a function that enables workflows to provision cloud and on-premise computing resources for training models.
• The freedom 🦅 to mix and match CML with your favorite data science tools and environments.

Note that CML does not include DVC and its dependencies (see the Setup DVC Action).

Note on v2

v1 of setup-cml was a wrapper around a set of npm installs. v2 installs CML from its pre-packaged binaries. Then attempts to run npm install --global canvas@2 vega@5 vega-cli@5 vega-lite@5 if you do not wish to install these tools pass vega: false to the action.

link to v1

Usage

This action is tested on ubuntu-latest, macos-latest and windows-latest.

Basic usage:

steps:
  - uses: actions/checkout@v3
  - uses: iterative/setup-cml@v2

A specific version can be pinned to your workflow.

steps:
  - uses: actions/checkout@v3
  - uses: iterative/setup-cml@v2
    with:
      version: 'v0.18.1'

Without vega tools

steps:
  - uses: actions/checkout@v3
  - uses: iterative/setup-cml@v2
    with:
      version: 'v0.20.0'
      vega: false

Inputs

The following inputs are supported.

• version - (optional) The version of CML to install (e.g. '0.18.1'). Defaults to latest for the most recent CML release.
• vega - (optional) Whether to install vega dependencies. Defaults to true. runs command npm install --global canvas@2 vega@5 vega-cli@5 vega-lite@5

A complete example

A sample CML report from a machine learning project displayed in a Pull Request.

Assume that we have a machine learning script, train.py which outputs an image plot.png:

steps:
  - uses: actions/checkout@v2
  - uses: iterative/setup-cml@v2
  - env:
      REPO_TOKEN: ${{ secrets.GITHUB_TOKEN }} # Can use the default token for most functions
    run: |
      python train.py --output plot.png

      echo 'My first CML report' > report.md
      echo '![](./plot.png)' >> report.md
      cml comment create --publish report.md

In general GitHub's runner token can be given enough permissions to perform most functions. When using the cml runner launch command a PAT is required

CML functions

CML provides several helper functions. See the docs.

Contributing

To get started after cloning the repo, run npm ci (clean-install). Before pushing changes or opening a PR run npm run format && npm run lint to ensure that the code is formatted and linted.

run npm run build to compile the action.