Tutorial: From Notebook to Kubeflow Pipelines to KFServing: the Data Science Odyssey
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Friday, 2020-11-20· at 15:10· Intrado Virtual Event Platform,·
summary
Deploys MiniKF in ~15 minutes, uses Jupyter + Kale converts notebooks to Kubeflow Pipelines without Docker builds, snapshots environment with Rok so notebook dependencies run unchanged, and logs everything to MLMD. Hyperparameter tuning uses Katib to run hundreds to thousands of parallel trials (demo shows a 50‑trial run and a larger live experiment), caches steps via PVC snapshots, then restores the best run into memory and KFServing exposes predictor + transformer for serverless inference.
AI-written summary · about
chapters (35)
- 0:00Introduction
- 0:23Agenda
- 1:12What is Kubeflow
- 1:42Use Cases
- 4:48User Interface
- 5:16Command Line
- 6:10ML Code
- 7:43Highlevel overview
- 9:01The agenda
- 9:35Deployment
- 10:10Creating a MiniKF
- 11:42Pipelines
- 12:16Benefits
- 12:49Workflow
- 14:03Beta Management
- 18:28Notebook Overview
- 20:04Annotations
- 20:42Compile Run
- 21:05Python Libraries
- 24:10Manual Pipelines
- 27:57Compile and Run
- 28:54Catib UI
- 29:41Experiments Graph
- 32:28Cached Steps
- 33:08Visualizing Steps
- 34:02KFServing
- 34:34Restoring a Notebook
- 35:21Restoring a Notebook with the State
- 36:43Restoring a Notebook with the Server
- 37:13Restoring a Notebook from a Snapshot
- 38:29KFServing API
- 39:13Preprocessing
- 41:34Models UI
- 46:01Summary
- 47:57Community
related talks
01h 27mTutorial: From Notebook to Kubeflow Pipel...
51mMachine Learning using Kubeflow and Kubernetes
46mAccelerating Machine Learning DevOps with Kubeflow
54mScaling Machine Learning Pipelines in Cloud
35mKubeflow: Data Scientist ready?
39mOperating Deep Learning Pipelines Anywhere Using Kubeflow
21mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)
36mIntroducing Kubeflow (w. Special Guests Tensorflow and Apache Spark)