Dagster + CML MLOps Template 🤖🔬🛠️

This repository is a minimal example of how to integrate CML (Continuous Machine Learning) and Dagster into your MLOps workflow. It uses GitHub Actions to orchestrate and automate a simple machine learning pipeline with reporting directly on your pull requests. 🚀🔁

Introduction 📄🚀

This project demonstrates how to implement a CI/CD pipeline using GitHub Actions that:

• Executes a Dagster job to train and evaluate a machine learning model (Random Forest).
• Uses CML to report metrics (e.g., accuracy) back to GitHub via PR comments.

By combining these tools, you get a reproducible, automated, and collaborative machine learning workflow in a fully Git-based environment. 🧪🛠️📊

What's Inside 🧬📁

• Dagster Pipeline: A simple ML job defined in Python using Dagster to:

  • Load and split the Iris dataset.
  • Train a RandomForestClassifier.
  • Evaluate the model’s accuracy.

• GitHub Actions Workflow: A CI/CD pipeline in .github/workflows/pipeline.yml that:

  • Sets up Python.
  • Installs dependencies.
  • Executes the Dagster job via CLI.
  • Logs accuracy to a file.
  • Posts a comment on the PR using CML.

Getting Started 🛠️🏁

To get this template running in your own repo:

1. Create a new repository using this template:

   • Click “Use this template” at the top right.
   • Name your new repository.

2. Clone your new repo locally:

   git clone https://github.com/yourusername/your-mlops-repo.git
   cd your-mlops-repo

3. Customize and test your workflow:

   • The GitHub Actions workflow is already set to trigger on pull_request with the main branch as a target.
   • You can manually trigger it by creating or updating a pull request with main as the target branch.

4. View the results:

   • Once the CI job completes, CML will post an accuracy score as a comment on the PR.
   • This ensures quick visibility into model performance without switching tools.

CML Reporting 📢📊

CML is used to post model evaluation results back to GitHub. Here's how it's done:

• During the GitHub Actions run, the Dagster job writes model accuracy to a metrics.txt file.
• CML reads this file and posts the contents as a PR comment using:

  cml comment create --file metrics.txt

You can expand on this by:

• Adding plots (e.g., confusion matrix).
• Tracking experiments with DVC or MLFlow.
• Exporting models or data as artifacts.

More info in the CML Docs.

License 📜

This project is licensed under the MIT License. Use, modify, and share freely!

Acknowledgments 🙏

• Huge thanks to the CML and Dagster communities.
• Inspiration from MLOps best practices and real-world workflows.