Make your data science workflows reproducible
summary
DVC versions datasets and models via metadata files (.dvc) and a content-hash cache so Git stores only small pointers while actual data lives in remotes; example uses a Vokras image set and Titanic pipeline. Pipelines are declared in dvc.yaml with params.yaml for test_size and random_seed, stages run with dvc repro and experiments listed with dvc exp show, and data pushed with dvc push. Reruns skip unchanged stages, CI/CD can trigger full reproducible runs.
AI-written summary · about
chapters (10)
- 0:00Introduction to reproducibility
- 1:11What are pipelines
- 2:42Common reproducibility issues
- 3:48Introducing DVC
- 4:54Data versioning example
- 8:17Restoring dataset versions
- 10:08Pipeline management demo
- 12:20Experiment tracking with DVC
- 14:12What DVC is and isn't
- 15:19Closing remarks
related talks
42mApplications in ML Drug Discovery pipelines
01h 32mHow to easily set up and version control your Machine Learning Pipelines
59mDevOps for Machine Learning Projects
46mFlexible ML Experiment Tracking System for Python Coders with DVC and Streamlit
33mHow to Track your Development Process with DVC
02h 10mBuild End-To-End MLOps Platform with Open Source DVC Ecosystem
38mUsing Reproducible Experiments To Create Better Models
27mLeveraging Git for ML Experiment Management