RAPIDS Workshop

3 hours of instruction

Introduces the open-source Python RAPIDS libraries for accelerating computation with GPUs (graphics processing units). Participants practice using the RAPIDS libraries for common ETL and machine learning workloads without having to program with low-level languages (e.g., C/C++).

PREREQUISITES

Participants should have prior experience using the Python language and, in particular, using standard Python tools for data analysis (notably NumPy, Pandas, Jupyter). No prior experience with GPU programming is required (although some prior exposure to Dask, while not mandatory, will be helpful).

LEARNING OBJECTIVES

1. ​Verify the availability of GPU hardware on a given system for accelerated performance.
2. ​Explain relevant GPU computing concepts in the context of data analysis pipelines.
3. ​Identify opportunities for GPU computation in existing Python data analysis pipelines.
4. ​Extend example Pandas/Scikit-Learn pipelines to scalable GPU-pipelines with RAPIDS.
5. ​Construct scalable machine learning pipelines in Python using RAPIDS from scratch.
6. ​Identify when GPU dataframes can benefit from using Dask-cuDF instead of cuDF for improved performance.