Hacker News Reader

Comment by tomp | Hacker News Reader

So... I checked the nbodies code.

Python: simple, short, very obvious what's going on, completely unoptimized. I wonder how much faster it would be, if it was (1) vectorized with numpy, and then (2) numba on top. I doubt it would be much less readable.

Julia: highly optimized (many "inline" annotations etc.), vectorized, much less readable than Python.

TL;DR: this isn't a fair comparison.

View on HN

The slower Julia variants don't seem less readable than the Python one to me and still are a lot faster.

That's the truth, the fact

I'd say most of those optimizations are not needed (`@inline`, `@ntuple`, `@inbounds` can also often be dropped).

Vectorization, in the sense of using `.+` etc, does not provide any speedup in Julia, it is only for notational convenience, and is at best the same speed as a loop, often slower.

> I'd say…

People do say, but without program measurements…

More than one Python program, more than one Julia program:

> many "inline" annotations etc.

https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

This one is still vectorized.

Is it "much less readable than Python"?

no, that it's not

I've seen lots of comparisons between Julia and Python, since the Julia community love those. So far I've seen zero using numpy.

generally numpy closes the gap to within 2x as long as all your operations are easily vectorizable, and none of your arrays have a size less than 100 or so. for small arrays, numpy is around 1000x slower than Julia because every operation has a large overhead (somewhere in the range of 10 us)

OK, so Julia beats stock Python for some benchmarks.

There is no shame in offering credit where due.

Does the "numpy community" provide those?

Reply on news.ycombinator.com