Developing and Deploying Models at Scale — without Spark!
summary
Live demo trains an h2o deep learning model on the Airline dataset (about 10 million rows) on a single X1 instance using 128 cores and terabytes of RAM, then publishes a Dockerized model and serves it as a restful API with autoscaling and sub-second responses; model training and deployment succeed onstage. The workflow uses Domino for execution, versioning and model management, and also shows teapot auto-ML evolving an SVM pipeline in ~2–3 minutes and deploying it as a service.
AI-written summary · about
chapters (8)
- 0:00Start
- 2:52Build and Deploy a Machine Learning Model
- 5:43Deploy a Model
- 6:26Live Demo
- 9:39Automated Version Control
- 12:09Serialized Docker Image
- 15:06Randy Olson
- 21:56Summative Assessment
related talks
33mProductionizing H2O Models with Apache Spark
27mDeploying distributed AI and machine learning in financial services
31mBeyond One Model: Scaling, Orchestrating & Monitoring
50mScalable Data Science & Deep Learning with H2O
17mKeynote: Scaling Deep Learning Models in Production Using Kubernetes
23mRethinking scalable machine learning with Spark ecosystem
43mDeploying Large Spark Models to production #bbuzz
39mOperating Deep Learning Pipelines Anywhere Using Kubeflow