Fast NetworkX and How Accelerated Backends Are Changing Graph Analytics
summary
NetworkX now dispatches hundreds of algorithms to accelerated backends so the same API runs on GPUs (cugraph + cuDF), GraphBLAS implementations, or falls back to pure-Python; conversion to GPU memory is cached so repeated calls avoid copy overhead. Empirical numbers: a between-centrality example drops from 7 minutes 44 seconds to 20 seconds, and a PageRank experiment reports trillion-edge PageRank in 19.3 seconds. Backend support is matured, documented, and shipping around the 3.4 cycle.
AI-written summary · about
chapters (12)
- 0:00Accelerated graph analytics
- 1:28Graphs everywhere today
- 5:24NetworkX as go-to library
- 7:54Backends supercharge NetworkX
- 10:03Faster graph backends needed
- 13:15Graphs in unexpected data
- 20:49Benchmarking NetworkX
- 23:37Data frames as graphs
- 24:45NetworkX backend ecosystem
- 26:09Q&A on backends
- 29:00Custom functions with backends
- 30:09Summary
related talks
36mNetworkX is Fast Now- Zero Code Change Acceleration
40mRun Large-Scale Graph Analytics Using The Most Popular Graph Analytics Library Available
32mZero Code Change GPU-Powered Graph Analytics with NetworkX and cuGraph
31mGraph-Based Machine Learning with Python
35mCUDA in Python- A New Era for GPU Acceleration
30mCUDA in Your Python: Effective Parallel Programming on the GPU
30mUnderstanding GPUs
37mGraphs, Networks and Python: The Power of Interconnection