Hacking Import for Speed: Writing a GPU Accelerator for Pandas

Explore the powerful technique of hacking Python's import system to accelerate pandas operations using GPU technology. Dive into an ambitious project that leverages cuDF, a GPU DataFrame library, to enhance performance without modifying existing code. Learn about import hacking basics, Pythonic proxy patterns, and how to apply these dynamic Python features to accelerate not only pandas but also third-party libraries. Discover the technical and social challenges encountered during the development process, and gain insights into potential solutions. Ideal for Python enthusiasts interested in the import system, performance optimization, and exploring lesser-known aspects of the language, this talk provides valuable knowledge for developers seeking to speed up the ecosystem built on numpy and pandas without code alterations.

Talks - Bradley Dice: Hacking `import` for speed: how we wrote a GPU accelerator for pandas