Accelerate Python* with NumPy & Other Smarter oneAPI Equivalents

Ignite performance of common Python* and pandas constructs by taking advantage of NumPy, SciPy, and pandas techniques powered by oneAPI.

This on-demand workshop details how to use key Intel® architecture innovations and libraries through a smart application of techniques that enable acceleration of 10x, 100x, or more, including:

• Detailed information on NumPy aggregations, universal functions, broadcasting, and other techniques.
• How to create outsized performance gains by replacing Python loop-centric or list-comprehensive applications with smarter equivalents that are more maintainable, efficient, and faster.

0:00 Introductions

1:35 Learning objectives

2:14 AI Tools

5:40 Intel® Developer Cloud

7:12 How to get started with the Intel Developer Cloud

22:08 Next step: follow the README

28:35 Poll: Were you able to complete the Git clone step?

33:55 NumPy: powered by oneAPI

36:05 Python* is great and fast

37:23 Python is slow

38:02 NumPy vectorization

38:55 Vectorization is not just theory

39:39 Why are these speedups so dramatic?

41:20 Comparing to simple loops in Python

43:00 Effect of noncontiguous memory access

44:38 Cache is used ineffectively and a real-life example

47:32 About memory

48:37 How to move code patterns to NumPy

49:15 Demonstration on how to create NumPy arrays

1:03:00 Quick resolution to fix broken code

1:03:43 NumPy universal functions

1:14:00 NumPy aggregation

1:19:00 NumPy broadcasting

1:25:00 NumPy where clause

1:35:00 NumPy with pandas

1:43:05 NumPy with SciPy

1:55:00 NumPy matrix multiplier

1:58:45 Call to action

2:00:20 Q&A