Speeding up Machine Learning tasks using GPUs in Python
summary
RAPIDS cuML implements GPU versions of Decision Trees and Random Forest with claimed 10x–50x speedup and an end-to-end demo on a 200 GB mortgage CSV running on DGX boxes; a separate Forest Inference Library (FIL) loads treelite/protobuf model bytes to run GPU inference for XGBoost, LightGBM, scikit-learn and cuML models. API mirrors pandas/scikit‑learn so code changes are minimal; installation via conda or Docker. Limitations: FIL currently targets binary classification and regression has some nondeterminism from parallel processing, so benefits show up only on larger datasets.
AI-written summary · about
chapters (13)
- 0:00GPUs in ML introduction
- 3:08Apache Arrow and GPU ETL
- 4:35RAPIDS and cuML
- 5:37Why GPU acceleration
- 7:45cuML algorithms overview
- 8:54Random Forest deep dive
- 12:27Multi-GPU Random Forest
- 13:53GPU inference with FIL
- 17:53Live demo vs scikit-learn
- 20:16cuML trade-offs
- 22:22RAPIDS release timeline and future plans
- 23:49Installation and getting started with RAPIDS
- 24:54Q&A
related talks
01h 12mGPU-Accelerated Data Analytics in Python
01h 08mGPU Data Science with Rapids
01h 36mRAPIDS: GPU-Accelerated Data Science for PyData Users
38mRAPIDS- Open Source GPU Data Science
41mAccelerating Data Science with RAPIDS
01h 22mHigh-Performance Data Science at Scale with RAPIDS, Dask, and GPUs
34mRAPIDS: How to accelerate your data science pipeline by orders of magnitude
16mGPU-Accelerated Data Science for PyData Users