Concurrent NumPy in Python

Concurrency in NumPy is not an afterthought

• Discover matrix multiplication that is 2.7x faster.
• Discover array initialization that is up to 3.2x faster.
• Discover sharing copied arrays that is up to 516.91x faster.

• NumPy has multithreaded algorithms and functions built-in (using BLAS).
• NumPy will release the infamous GIL so Python threads can run in parallel.
• NumPy arrays can be shared efficiently between Python processes using shared memory.

• Parallel array multiplication, common math functions, matrix solvers, and decompositions.
• Parallel array filling and parallel creation of arrays of random numbers.
• Parallel element-wise array arithmetic and common array math functions
• Parallel programs for working with many NumPy arrays with thread and process pools.
• Efficiently share arrays directly, and copies of arrays between Python processes.

• The importance of concurrency when using NumPy and the cost of approaching it naively.
• How to perform common NumPy operations and math functions.
• How to install, query, and configure BLAS libraries for built-in multithreaded NumPy functions.
• How to use Python concurrency APIs including threading, multiprocessing, and pools of workers.