Workshop: Intro to GPU data science with RAPIDS
summary
Shows RAPIDS running on a Tesla V100 processing a 7 GB Twitter CSV with cuDF DataFrame and cuML random forest: preprocessing converts float64 to float32 to get GPU speed, uses set_allocator managed to avoid memory oversubscription, and times classification at ~726 ms versus minutes for scikit-learn on CPU. Finds suspicious accounts with average daily statuses up to ~1.2M that are later banned. Notes Colab limits (CUDA/cuDF 0.14), BlazingSQL for fast I/O, and nightly 0.15 features (cosine similarity) that further accelerate searches.
AI-written summary · about
chapters (9)
- 0:00Introduction to GPU data science
- 3:09Why GPUs matter today
- 13:24GPU architecture and gaming history
- 17:59From hardware to RAPIDS
- 21:12Introducing RAPIDS libraries
- 26:11Setting up RAPIDS environment
- 31:51Twitter data analysis with RAPIDS
- 43:30Machine learning with RAPIDS
- 54:55Q&A on GPUs and RAPIDS
related talks
41mAccelerating Data Science with RAPIDS
01h 12mGPU-Accelerated Data Analytics in Python
38mRAPIDS- Open Source GPU Data Science
01h 08mGPU Data Science with Rapids
39mAccelerating Data Science with RAPIDS
01h 22mHigh-Performance Data Science at Scale with RAPIDS, Dask, and GPUs
01h 36mRAPIDS: GPU-Accelerated Data Science for PyData Users
16mGPU-Accelerated Data Science for PyData Users