KEYNOTE: Scaling Out PyData
summary
Claims that Python now scales both up and out with concrete tools and numbers: numba compiles Python to native code with GPU targets and reports up to 2000x speedups in some cases, linear-algebra work shows 10–100x gains over PySpark, and combining libraries yields sub‑second rendering of three billion points via DataShader and Bokeh. Anaconda bundles MKL for fast BLAS, Conda Forge supplies reproducible packages, Dask runs interactive pandas/numpy-like computations across hundreds of machines, Blaze remains a backend abstraction, and Data Fabric is an alpha for cross-language data access.
AI-written summary · about
chapters (8)
- 0:00PyData introduction and growth
- 3:37Python ecosystem for scalable computing
- 10:00Anaconda as data science platform
- 12:49Numba and Dask for scale-up and scale-out
- 18:52Blaze: data expressions and runtime
- 21:02Numba JIT compilation and performance
- 28:17Dask for parallel computing
- 36:45Conclusion: Python scalability
related talks
45mPython for Speed, Scale, and Science
14mKeynote: Scalable Python We Can Afford
38mA Hitchhiker's Guide to Data Science
38mKeynote: Python for Pythonistas
26mBodo: Supercomputing-Like Performance and Scale for Python/Pandas
38mState of Data Science & Machine Learning
30mEmpowered by Python - A success story
23mData Science at Scale Using dask and Numba