Automate the boring stuff with MLOps | Data Days 2022

TL;DR

Casavo's team showcases tools for automating machine learning with a focus on data versioning, hardware training, configuration, deployment, and cloud training.

Transcript

let's continue you guys are mic'd up yeah perfect hello um let's continue because you have numerous um opportunities to chat later on the boats so we want to keep it going we want to continue automating the boring stuff we just learned a lot about planting seeds and automating away dump decisions and we want to continue provide inspiration to you a... Read More

Key Insights

• 🔨 Automation tools like poetry, Hydra, and BentoML streamline ML development and deployment.
• 🦻 MLflow aids in tracking training and managing ML models effectively.
• 🔨 Configuration management with tools like Hydra simplifies complex application setups.
• 😶‍🌫️ Leveraging cloud training tools like CML enhances scalability and efficiency in training ML models.
• 📽️ Standardizing ML processes through reproducibility and efficient training pipelines is crucial for successful ML projects.
• 🔨 Challenges in ML deployment to production are addressed by tools like BentoML, simplifying the process.
• 😤 Collaboration within the team and leveraging shared practices enhance efficiency and accuracy in ML projects.

Questions & Answers

Q: How did Casavo's team address the challenge of standardizing ML processes?

Casavo's team leveraged tools like poetry for environment setup, data version control, and Torch lightning for efficient training to standardize their ML approach.

Q: What role does Hydra play in Casavo's ML pipeline, and how does it simplify configuration management?

Hydra is used to elegantly configure complex ML applications, allowing easy combination of configurations, directory management for artifacts, and running sweeps for hyperparameter tuning.

Q: How does MLflow benefit Casavo's team in tracking and managing ML training?

MLflow serves as a solution for tracking trainings, managing artifacts, and even acting as a model registry, enabling smoother ML pipeline and reproducibility of experiments.

Q: What challenges did Casavo face in deploying ML models to production, and how did BentoML address these challenges?

Casavo encountered deployment complexities, but BentoML simplified the process by automating dependency management, providing endpoint documentation, and creating Docker images for seamless deployment.

Summary & Key Takeaways

• Casavo's team discusses their journey of automating machine learning tasks to standardize processes and avoid reinventing the wheel.

• Tools like poetry, data version control, Torch lightning, Hydra, MLflow, and BentoML are highlighted for streamlined ML development.

• They emphasize the importance of reproducibility, efficient training, configuration management, and seamless deployment in their ML projects.