Accelerate Model Hyperparameter Search with RAPIDS and SageMaker
summary
Shows a reproducible workflow that containers RAPIDS (cuDF, cuML), Dask and XGBoost into an AWS SageMaker estimator, runs hyperparameter search over an airline flight‑delay dataset, and deploys the best model for real‑time inference. Benchmarks claim ~12x wall‑clock speedup for large HPO (100 jobs) with GPUs vs CPUs, ~4.5x cost reduction, and a 70% spot savings example; a small demo runs 10 jobs (2 parallel) finishing a sanity run in 19s with ~91.5% score. RAPIDS cuDF & cuML and Dask multi‑GPU are the used tools and the pipeline works end‑to‑end.
AI-written summary · about
chapters (8)
- 0:00Start
- 8:28install with conda or with a container
- 41:05compute platform
- 1:02:20get a live environment with a rapids kernel in your notebook
- 1:13:56split our training
- 1:17:10run this on multiple gpus
- 1:17:42creating a local cuda cluster
- 1:21:06drop the samples with missing values
related talks
01h 22mHigh-Performance Data Science at Scale with RAPIDS, Dask, and GPUs
41mAccelerating Data Science with RAPIDS
34mRAPIDS: How to accelerate your data science pipeline by orders of magnitude
38mRAPIDS- Open Source GPU Data Science
39mAccelerating Data Science with RAPIDS
01h 12mGPU-Accelerated Data Analytics in Python
01h 36mRAPIDS: GPU-Accelerated Data Science for PyData Users
20mEnd-to-End Data Science on GPUs