Introducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
summary
Kubeflow packages a set of interoperable components to move models from laptop to cluster and into production, combining JupyterHub for notebooks, TensorFlow Serving for inference and Argo pipelines for repeatable, multi-step experiments. It supports PySpark (Arrow improves serialization), Katib for hyperparameter tuning, and cross-cloud deployment but currently has API churn (0.5 → 0.6) and limited built-in data versioning, so it works for reproducible training and serving but requires significant ops effort and careful handling of data gravity.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:56Trevor Grant
- 4:33History of Predictive Analytics
- 9:31Components Buffet
- 10:10How does this fit in to the Ecosystem
- 18:55A quick detour into PySpark's internals
- 30:18Questions? Half-baked Demo?
related talks
36mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
21mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
31mKubeflow - Data Science on Steroids
35mKubeflow: Data Scientist ready?
14mKubeflow: Where ML Meets Kubernetes
35mBuilding Cross-Cloud ML Pipelines with Kubeflow with Spark & Tensorflow
46mKubeflow, TensorFlow on Kubernetes
39mOperating Deep Learning Pipelines Anywhere Using Kubeflow