Introduction to CML

CML (Continuous Machine Learning) is an open-source library that helps automate Machine Learning (ML) workflows using Git’s version control system. It bridges the gap between DevOps and ML, providing seamless integration of CI/CD pipelines for ML models.

Setting Up CML

To get started with CML, you need to install it via npm or yarn:

 npm install -g cml # or yarn global add cml 

Common CML API Examples

cml-runner

The cml-runner command creates a cloud runner to execute a job.

 cml-runner \
  --repo https://github.com/your/repo \
  --token $GITHUB_TOKEN \
  --cloud aws \
  --cloud-region us-east-1

cml-send-comment

The cml-send-comment command posts a comment on a specific commit, pull request, or merge request.

 cml-send-comment \
  --token $GITHUB_TOKEN \
  --repo https://github.com/your/repo \
  --pr \
  --publish evaluation_report.html

cml-publish

The cml-publish command uploads an asset (e.g., text, image) to the remote storage and returns a publicly accessible URL. This is useful for sharing results within the CI workflow.

 cml-publish metrics.png --md 

Practical Application Example

Let’s dive into a simple application that uses these CML APIs. We’ll create a CI pipeline to automate ML experiment results reporting.

Setting Up the Pipeline

Create a .github/workflows/cml.yaml file:

 name: ML Experiment
on: [push]
jobs:
  run:
    runs-on: [self-hosted, cml]
    container: docker://dvcorg/cml:0-dvc2-base1
    steps:
      - name: Checkout
        uses: actions/checkout@v2
      - name: Setup CML runner
        run: |
          cml-runner --cloud aws --cloud-region us-west --repo ${{ github.repository }} --token ${{ secrets.GITHUB_TOKEN }} --idle-timeout=300
      - name: Train model
        run: |
          python train.py
      - name: Evaluate model
        run: |
          python evaluate.py
          cml-send-comment --repo ${{ github.repository }} --token ${{ secrets.GITHUB_TOKEN }} --pr --publish evaluation_report.html

This GitHub workflow sets up a CML runner, trains a machine learning model, evaluates the model, and posts the evaluation results as a comment on the pull request.

Conclusion

CML simplifies the integration of CI/CD practices in ML workflows, making model iteration and deployment seamless. Using CML’s API commands like cml-runner, cml-send-comment, and cml-publish, you can build robust and automated ML pipelines that enhance collaboration and productivity.

Hash: 737f59eac62477359fc3819b35fd5ccc71b60320a332a28073d01eecfb8f2e90