Post ApT6CyojTdNVdgMP1E by kjr@babka.social
(DIR) More posts by kjr@babka.social
(DIR) Post #ApT4MC5JJoqHU8NeGu by kjr@babka.social
2024-12-27T10:15:18Z
0 likes, 0 repeats
Nice article but... honestly... I never understood why people uses Conda/Anaconda. Which are the advantages of it?Saying Goodbye to Anaconda?#python #datascience https://robert-mcdermott.medium.com/saying-goodbye-to-anaconda-91c18ddf89bb
(DIR) Post #ApT4VB36XR2MsadFfE by paul@digitalstuntfactory.com
2024-12-27T10:16:54Z
0 likes, 0 repeats
@kjr um, I use it for package management and stuff. That's about my technical insight into its benefits.Why wouldn't I want to use it? 🤔
(DIR) Post #ApT4lTnoa6WdfwFoh6 by raf@babka.social
2024-12-27T10:19:50Z
0 likes, 0 repeats
@kjr For things that depend on c or c++ libraries, it's a much easier package management system than pip or some mutant combination of system package management and pip that never seems to install fully correct. Lots of machine learning libraries still have the best chance of installing if you use conda vs other optionsAlso back when it was introduced, there weren't these binary wheel packages.
(DIR) Post #ApT5GJFtqRTPwR5jHc by kjr@babka.social
2024-12-27T10:25:26Z
0 likes, 0 repeats
@paul There are several reasons.First of all is a kind of "take all", very heavy and you have to install the full library on the machine. Is something I don't want to do every time I configure a virtual machine for a task: waste of time and money.Second, package management... if you are using their packages is OK, but my experience building a library and trying not to go crazy with the dependencies, it can be a challenge. There are other minor reasons, as problems with versioning in cases in which you have to import same library in different versions for different modules in the same project, without duplicity in virtual environments... but this problem maybe is a mistake of us, I have not contrasted it with other teams.
(DIR) Post #ApT5K8jst1HOU3wWQK by ide@masto.ai
2024-12-27T10:26:08Z
0 likes, 0 repeats
@kjr Conda is nice because it handles everything seamlessly and more. It replaces pyenv, venv, poetry/etc. and even the system package manager for non-python packages in many cases. UV seems nice, but still lacks feature parity with Conda due to lacking support for non-python dependecies.
(DIR) Post #ApT5MuerhBSpgIuwTo by ide@masto.ai
2024-12-27T10:26:38Z
0 likes, 0 repeats
@kjr Conda is nice because it handles everything and more seamlessly. It replaces pyenv, venv, poetry/etc. and even the system package manager for non-python packages in many cases. UV seems nice, but still lacks feature parity with Conda due to lacking support for non-python dependecies.
(DIR) Post #ApT5PH3NOvozibVZxI by kjr@babka.social
2024-12-27T10:27:02Z
0 likes, 0 repeats
@raf "Lots of machine learning libraries still have the best chance of installing if you use conda vs other options"Which for instance? I have heard it, but I never saw the issue.
(DIR) Post #ApT5puDkOu72is3fpw by raf@babka.social
2024-12-27T10:31:50Z
0 likes, 0 repeats
@kjr scikit-learn, astropy, gdal, mlpack, pymc, gensim, manim, jax, and pytorch are a few that come to mind that I've seen people have struggles installing due to those dependencies
(DIR) Post #ApT62jrcDdNWMjpEqu by kjr@babka.social
2024-12-27T10:34:09Z
0 likes, 0 repeats
@raf I don't know what to say. Gensim I used earlier, never had problems.Scikit-learn and pytorch... I had the opposite experience, in a concrete project problems with Conda (not heavy problems). I use them often and I don't have any problem.
(DIR) Post #ApT6CyojTdNVdgMP1E by kjr@babka.social
2024-12-27T10:36:00Z
0 likes, 0 repeats
@raf I see.... Apple is a complete diverse universe for me, I cannot say a lot about.Usually I work in Linux powered virtual machines.
(DIR) Post #ApT6lzn1qwzAn5GCH2 by paul@digitalstuntfactory.com
2024-12-27T10:42:22Z
0 likes, 0 repeats
@kjr got it. For what I do, I have a very much not recommended approach for virtual environments. I mostly just keep using the same virtual environment for all of my projects (these are mostly personal projects).I could probably just get by with a single Python install and packages installed via pip or pipx. 🤔
(DIR) Post #ApT7ebiV3dGU3ZBOJE by raf@babka.social
2024-12-27T10:52:11Z
0 likes, 0 repeats
@kjr None of this is to say Conda is necessarily the correct solution in 2024. But the benefits, especially when it was first released were legit. Package management on python, especially in scientific-python land remains a shitshow