KubeCon US 2019
Creating reproducible machine learning system is hard! It is because challenges pertaining to reproducibility stem from all 5 ingredients of machine learning systems - software, hardware, algorithms, process and practice and data.
In this talk, I talk about how we can write 100% reproducible ML code. Extend this discussion to write end-to-end reproducible ML workflow using Pachyderm, Kubeflow, Katib, ModelDB, Seldon and Argo.
