Julien Jerphanion • 12/19/2021

Performance and scikit-learn (4/4)

This technical article details performance experiments on the PairwiseDistancesReductions design in scikit-learn. It focuses on the PairwiseDistancesArgKmin implementation, examining hardware scalability (thread/core usage) and computational efficiency for Euclidean distance calculations. The analysis includes benchmarks, discussions of Amdahl's law, and uses Linux `perf` tools to profile CPU cycles and cache behavior in the k-nearest neighbors algorithm.

0 comments

#performance optimization #Scikit Learn #Parallel Computing