Vectorization Techniques for 1000x Python Speedups - No C or Spark Required

Discover how to achieve 1000x speedups in Python without resorting to C or Spark in this 41-minute conference talk from EuroPython 2024. Explore vectorization techniques for optimizing multivariate Python functions using matrices, library knowledge, and creative problem-solving. Learn three core tricks: converting conditional logic to set theory, stacking vectors into matrices, and shaping data to match library expectations. Examine real-world examples from Bloomberg's ESG Scores, including time-series computations and complex financial models with numerous if/else branches. Gain insights into rewriting pandas backfill operations and simplifying cases using De Morgan's laws and sparse matrix representations. Conclude with an overview of cutting-edge tools and acquire a concrete strategy for vectorizing financial models to dramatically improve performance.

How we used vectorization for 1000x Python speedups (no C or Spark needed!)