Vectorization: How slow Python runs fast code

on YouTube ↗35m

" summary

Vectorization means using chunked numeric representations and bulk native routines so Python code runs C/Rust-speed on large arrays. Using numpy.sum on a 30 million–element array measures about 100x faster than a Python loop (loop ~1.5s, numpy ~0.03s in the demo). Watch for tradeoffs: creating temporaries (abs(array) then mean) doubles memory; use Numba for in-place loops or avoid per-item callbacks like pandas.map which reintroduce Python overhead.

AI-written summary · about

" chapters (22)

1. 0:00Intro
2. 0:13Why software speed matters
3. 1:36Feedback cycle time matters
4. 2:27The paradox of Python performance
5. 2:59Arrays: a shared, fundamental data structure
6. 4:12The vectorization solution
7. 5:17An example: Same results...
8. 7:01An example: ...different speed
9. 8:39Talk is about CPython
10. 11:32Iteration in CPython: slow!
11. 12:25Doing math in Python: slow!
12. 14:59Memory layout of C arrays
13. 15:59Memory layout of Python lists
14. 16:55Carray vs. Python list
15. 17:47NumPy arrays: a better memory layout for numeric data
16. 19:19Example: Summing one million integers
17. 22:51What does "vectorization" mean?
18. 24:47Digression: Single Instruction, Multiple Data (SIMD)
19. 27:10Avoiding built-in Python objects: example
20. 28:18Avoid iterating in Python
21. 31:16Avoid calling back into Python
22. 33:29Summary: Writing fast data processing code

" related talks

• 29mVectorize using linear algebra and NumPy to make your Python code fastJodie Burchell · PyCon US 2023
• 34mNumPy: vectorize your brainEkaterina Tuzova · EuroPython 2015
• 01h 31mMaking Computation Easier with Cool Numpy TricksKirit Thadaka · PyData Los Angeles 2018
• 28mWhy Python Loops Are "Slow": A Gentle Introduction to Interpretation, Bytecode, and Fast CodeTyler Riccio · PyOhio 2025
• 23mPerformance Python: Seven Strategies for Optimizing Your Numerical CodeJake VanderPlas · PyCon 2018
• 01h 27mMaking computation easier with cool Numpy tricksKirit Thadaka · PyData London 2018
• 40mSpeed Without DragSaul Diez-Guerra · PyData Berlin 2014
• 37mUsing Numba Effectively TodayCheuk Ting Ho · PyData NYC 2022