Perhaps for engineers, but for the rest of the scientific community it's pretty obvious: Rust doesn't have nearly the coverage that NumPy does. A quick check, I couldn't find a GMRES solver in Rust, which is extremely useful for solving large linear systems and hardly an obscure algorithm.

FWIW this is the same situation NumPy was in a while back ago, but instead of boldly asserting that what they had was enough, they looked to other prevalent languages to figure out what was missing and packages like pandas and statsmodels came out.

As someone who likes rust, python is just easier to code for exploration and the kinds of advantages you get for using rust aren’t particularly relevant.

The entire surrounding ecosystem. Which includes not just the core numpy/scipy libraries but the stats, modeling, ML, plotting, etc. Plus the wealth of other Python stuff it can integrate with (I work at an almost entirely Python shop -- data science people work in the same language as pipeline and application people, which is really an overlooked thing). Plus the ease of installation and management from Anaconda. Plus the notebook format for easy sharing. Plus... well, lots and lots and lots of things. "We have a fast numerical library" is step one of about ten thousand to achieving parity with what the Python/numpy/scipy ecosystem does.

They don’t compete since you could reimplement your compute kernels in Rust and use them directly or as (g)ufuncs with the numpy container, used by the whole science and data stacks (visualization stats etc). Numpy compute APIs are for prototyping or compute which isn’t CPU bottleneck.

So for the majority of numpy users Rust competes with the likes of Numba, Cython, etc, which are hard to beat if you’re already coding in Python.

Numpy is great because of all the frameworks built on top. What's the story with Rust?

Numerical computation is just one part of scientific code. Linear Algebra, data analysis, modelling, and visualization, are equally important.

TensorFlow doesn’t have a Rust interface.

Isn't this a Rust interface for TensorFlow?

https://github.com/tensorflow/rust

I avoid rust due to the annoying, symbol-heavy syntax that requires too much Shift-key usage.