Breaking the Speed Limit: Fast Statistical Models with Python 3.14...
summary
Concrete claim: Python 3.14 free‑threaded build plus an experimental JIT, Numba, and JAX let Python reach C‑like throughput on real statistical workloads. Two reproducible workloads are implemented and validated: k‑means (Numba wins on small K/D loop‑heavy cases, NumPy wins for dense BLAS shapes) and a permutation test (GPU gives 8.5× GPU speedup at R=5,000 and P=500,000). Optimized kernels match NumPy within 3.1e‑14 and label agreement 1.0. Decision flow is simulate and validate, identify bottleneck, choose tool: Python 3.14 free‑threaded, Numba, JAX.
AI-written summary · about
chapters (16)
- 0:00Optimizing statistical models
- 1:09Validation in statistical code
- 4:41Python tools for readable code
- 6:47K-means clustering test case
- 10:20Validating K-means before optimization
- 13:38Optimizing K-means with NumPy, Numba, JAX
- 16:45Choosing tools by bottlenecks
- 17:05Permutation tests challenges
- 18:53Optimizing permutation tests with batching
- 20:45Validating permutation test results
- 21:51Permutation tests scaling
- 22:35GPUs vs CPUs in permutation tests
- 23:40Parallelism limits and scaling
- 25:46Choosing computational engines
- 28:17AI coding tools and ownership
- 29:01Summary
related talks
27mSpeed up Python with hardware accelerators
47mBig ideas shaping scientific Python: the quest for performance and usability
44mPython's Bright Future in Science
01h 33mPower up your work with compiling and profiling
55mAccelerated Python: The Community and Ecosystem
41mSwift + Metal: The future of scientific programming?
02h 55mGPU Accelerated Python w: CUDA
39mPushing Cython to its Limits in Scikit-learn