DevOps for Science: using continuous integration
summary
Continuous integration extends to ML by versioning data, running training on demand, and surfacing detailed metric reports in pull requests; the team builds an open source toolkit—CML—that integrates with GitHub Actions and GitLab CI, uses DVC metrics diff and plots to compare branches, and can provision cloud GPUs via docker machine (moving toward Terraform). Reports embed markdown, images and tensorboard links in PRs. The repo shows active examples and the video series gets almost 17,000 views.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:07ML workflow challenges
- 4:41DevOps principles for machine learning
- 5:44Continuous integration
- 7:51ML workflow differences
- 10:59CI/CD for ML
- 12:24Continuous Machine Learning
- 13:48GitHub Actions & GitLab CI
- 15:54ML CI/CD priorities
- 16:16GitHub Actions setup
- 18:43Metric reports in PRs
- 21:07CML report flexibility
- 22:52Reviewing data & models
- 24:15Versioning data with DVC
- 26:04Consequences for continuous integration
- 26:48Cloud resources for training
- 28:14Summary
related talks
56mDevOps for Machine Learning
59mDevOps for Machine Learning Projects
29mDevOps for Data Science? - automate the boring stuff and leverage the OSS ecosystem
17mContinuous deployment for ML: The new software development lifecycle
24mFrom Model to Production: Automating the ML Lifecycle with CI/CD
57mDevOps for Machine Learning
02h 10mBuild End-To-End MLOps Platform with Open Source DVC Ecosystem
26mApplying DevOps practices in Data and ML Engineering