Tutorial: Hyperparameter Optimization (HPO) with RAPIDS on AWS SageMaker

rapids.ai YouTube Playlist — Getting Started Tutorials and Examples

This video tutorial will show you how to combine RAPIDS and Amazon SageMaker to accelerate hyperparameter optimization (HPO) and find the best version of your model before serving it to the world. HPO can take an exceedingly long time on a non-accelerated platform. When we combine the power of GPU acceleration within a node using RAPIDS and parallel experiments running across multiple nodes using AWS SageMaker, we can get impressive results. For instance, we find a 12x speedup in wall clock time and a 4.5x reduction in cost when comparing GPU and CPU EC2 Spot instances on 100 XGBoost HPO trials using 10 parallel workers on 10 years of the Airline Dataset hosted in an S3 bucket.

Learn how to spin up a SageMaker instance and quickly launch a demo notebook from the official sagemaker-examples repository.

We’ve built in lots of flexibility into the workflow. We invite you to explore the many configuration options as well as to plug in your dataset.

RAPIDS HPO now integrated into the SageMaker Examples UI

Tutorial Chapters:

1 — Introduction and Concept Overview

2 — Spinning up a SageMaker Instance

3 — Notebook Walkthrough with Description of Key Choices

For more details on getting started, check out the code repo as well as the RAPIDS cloud & RAPIDS HPO webpages. Find us on Slack or file a GitHub issue with suggestions.