←back to thread

I don't like NumPy

(dynomight.net)
480 points MinimalAction | 1 comments | | HN request time: 0.304s | source
1. frollogaston ◴[] No.44000171[source]
About optimization, I feel like NumPy is meant to be a de facto standard and reference implementation. It covers all use cases with decent efficiency, not the fastest way possible. There are more limited drop-in replacements that use more CPU parallelism or GPU if NumPy isn't fast enough for your use case. Just wish it were clearer which NumPy build I'm installing, cause apparently `pip3 install numpy` on my Mac gave me something built with the worst flags possible.

About >2 dimensions, I always found this confusing in NumPy but just chalked it up to >2 dim arrays being inherently confusing. Maybe there really is a better way.