Experimentación ágil de machine learning con DVC
1 like200 views
The document discusses the importance of version control in machine learning projects, highlighting issues like dataset evolution and reproducibility crises. It introduces DVC (Data Version Control) as a solution for versioning data and models, tracking experiments, and managing pipelines effectively. The document also provides links to additional resources for learning more about DVC and engaging with the community.
Experimentación ágil de machine learning con DVC
- 1. Version Control for Machine Learning Projects @mroutis
- 2. About me - Name: Ramón Valles - Handle: mroutis - Work: Software developer - Company: iterative.ai @mroutis
- 4. Why versioning is important? @mroutis
- 5. Datasets evolve over time @mroutis
- 7. “Yea, we have version control” @mroutis https://www.reddit.com/r/ProgrammerHumor/comments/63rcvc/yea_we_have_version_control/
- 8. How do we identify each version? @mroutis
- 9. https://www.reddit.com/r/ProgrammerHumor/comments/99a9k8/version_control @mroutis
- 10. Manual work is error prone. @mroutis
- 11. Is versioning data / models enough? @mroutis
- 13. Reproducibility Crisis “I was recently chatting to a friend whose startup’s machine learning models were so disorganized... training the same model get different results!” - Peter Warden (2018) https://petewarden.com/2018/03/19/the-machine-learning-reproducibility-crisis/ @mroutis
- 14. @mroutis
- 15. What have been done so far to address this? @mroutis
- 17. Manual work is error prone. @mroutis
- 19. - Chain processes together - Reproduce in the correct order - Make it easier to return to old work - The output of one stage is the input of the next on @mroutis Pipelines
- 20. How does DVC address this problems? @mroutis
- 21. @mroutis
- 24. @mroutis
- 28. Want to learn more? - Check out docs: https://dvc.org/doc - Chat with us: https://dvc.org/chat - Contribute on GitHub: https://github.com/iterative/dvc - Follow us on Twitter: @DVCorg @mroutis
- 29. Thanks for listening! @mroutis Thanks for Listening - Chris Thile
- 30. Thanks for the invitation!