Hacker News Reader

Comment by anigbrowl | Hacker News Reader

I've updated it based on your suggestion.

I originally left off a description because NetworkX is the library for doing graph things in Python and I kind of assumed people would know. While it's not the most performant or cutting edge, it's so well-established that everyone else aims for interface compatibility with it, much like pandas and tabular data.

It's pretty neat that they've decided to leverage this by opening up the backend, allowing more performant/application specific offerings like SNAP or CUgraph to do the heavy lifting.

View on HN

> I kind of assumed people would know

Not everyone writes Python, and not all Python programmers have heard of this particular library.

Not all people are programmers and not all people have heard of programming. This post is catered to an audience that probably knows what NetworkX is.

I've never heard of it.

I have written Python for 10 years on and off and have never come across it, or if I did I had forgotten about it.

This is not the same thing. Hacker News as a website is catered to an audience that is mostly familiar with programming. I would say it's safe to assume for any given visitor that they are probably aware of basic computing and programming terminology, but not that they would know what "NetworkX" means.

The new title is fitting (it specifies what it's a Python library now).

It was my go-to package in python, when I had to build graphs and ascertain acyclicity, single source, shortest path. They weren't huge graphs (at most 100s of elements), so networkx certainly did the work fast enough.

"the library"?

Igraph and graph tool might have something to retort to that. Up until this update, they were far far faster too (and still likely are if you ignore the ability in this update to specify different backends)

Yep we use cuGraph a ton, and stick to the CPU stuff for < 100k graphs, esp for their more niche algs. Exciting days!

Yes. I noted that it's not the most performant, indeed it's famous for being slow. But it's so popular, has such extensive code coverage, and such a simple API that it's gained default status, not unlike how Python is the default language for data science despite its limitations.

I always used IGraph because there are both R and Python versions, and I learned R first. I never had any complaints about it for the baby-scale problems I used it for, so I never bothered to try NetworkX.

Reply on news.ycombinator.com