Best distro for data science? [request]
Ubuntu seems like it has the best compatibility, but any other suggestions for data wrangling, data analysis, data visualization, and machine learning in Julia, Python and R?
Syndicated from the fediverse. Read and engage on the original instance.
View original on infosec.pubView original on infosec.pub
16 replies
Any distro you are comfortable with.
Which is the correct answer to 99% of questions that start with "Which distro..."
That's awesome! I hadn't heard about COSMIC DE.
Well put. The one thing I would add is using the Nix package manager on a distro other than NixOS! I'm daily driving Fedora 39 + Nix (home-manager) with zero problems. My pick would either be Fedora or Debian.
Tons of good documentation either way. Flatpak the packages you, no kidding, need to be easy / consistent to debug. Non-root podman for containers. Nix for more up to date packages than are available in the native repos (especially useful with Debian) + the other benefits like nix-shell.
Just use any distro you like and install the packages you need. Done.
This is going to be unpopular, but you can easily compile both Python and R and configure them to your liking. For Python you can even use Anaconda3 and forget about installing most packages by yourself.
As for Julia, I usually just install the precompiled binary package.
So, any distribution you feel comfortable with will do.
Learn docker on the distro you're most comfortable with.
Stick to ubuntu or red hat alts. They are whats gonna be in any cloud or HPC cluster.
For data science, it depends on what GPU you plan to use. If it's an Nvidia brand GPU, go with Ubuntu or Fedora. I say from personal experience that it is easier to get Nvidia drivers working on Ubuntu or Fedora than on most other distros I have tried. If it is a Radeon GPU, it will work fine on pretty much any distro at all since Radeon does a good job following Linux standard APIs for graphics card drivers, so for Radeon products I would also recommend Debian or Mint (along side Fedora and Ubuntu).
Debian got me through grad school.
Not the latest and greatest (if you run stable), but if you need the latest e.g. Julia, it's not too bad to compile it.
Use anything you want. All distros should support those packages, use what you're the most comfortable with.
I personally would recommend Fedora Silverblue/ it's other atomic variants or uBlue especially.
It's pretty much unbreakable, modern and supports ALL distros' package managers through Distrobox. It's also pretty simple in my opinion, since you pretty much don't have to worry about traditional package management.
I think you're searching something reliable and simple, so this would be a solid choice.
Mint would be great too
We use EL (Specifically Rocky, a rebuild of Redhat) for this, but I strongly suspect that any of the main distros will be absolutely fine provided they have modern enough versions of the software you need.
Have you considered something like jupyterhub?
I would say Nix but I presume you are a beginner so perhaps stick with Pop OS, Mint or Ubuntu.
I've been running Linux since the 90s and use popos. Nothing "beginner" about it.
It is beginner friendly, that was my point. At the end of the day it is Linux and you can do everything you can do on a normal Linux too.
R-ch (pronounced Arch btw)