I think this really crystallised for me why I avoid conda whenever I can. In trying to solve everyone's problems it became a system where if you want to do things outside of conda's wheelhouse...
I think this really crystallised for me why I avoid conda whenever I can. In trying to solve everyone's problems it became a system where if you want to do things outside of conda's wheelhouse (hidden python packaging joke there?) you run the risk of disturbing the system.
Conda envs just feel fickle whenever I use them, though I do definitely admit that sometimes they simplify the installation for packages with lots of external dependencies.
I vibe with this. The lack of real first-party support for tools like poetry, conda, and the other tooling the author mentions leads to many cases where they break down and end up as a time sink....
I vibe with this. The lack of real first-party support for tools like poetry, conda, and the other tooling the author mentions leads to many cases where they break down and end up as a time sink. I've moved back to simpler patterns like those suggested in this article. The hassle and inconsistency in Python's packaging ecosystem needs a lot of work. I still prototype things in Python, but I have more often been translating prototypes, once more established, into Rust projects which just don't break in the ways Python projects do. I'm also getting excited for some of the newer languages that are gearing up to replace (or improve on) Python in the ML and DL space, and I hope the communities can evolve. Perhaps the Python ecosystem can improve here over time, but I'm concerned it'll be a low priority and slow going even if it were a priority.
I think this really crystallised for me why I avoid conda whenever I can. In trying to solve everyone's problems it became a system where if you want to do things outside of conda's wheelhouse (hidden python packaging joke there?) you run the risk of disturbing the system.
Conda envs just feel fickle whenever I use them, though I do definitely admit that sometimes they simplify the installation for packages with lots of external dependencies.
I vibe with this. The lack of real first-party support for tools like poetry, conda, and the other tooling the author mentions leads to many cases where they break down and end up as a time sink. I've moved back to simpler patterns like those suggested in this article. The hassle and inconsistency in Python's packaging ecosystem needs a lot of work. I still prototype things in Python, but I have more often been translating prototypes, once more established, into Rust projects which just don't break in the ways Python projects do. I'm also getting excited for some of the newer languages that are gearing up to replace (or improve on) Python in the ML and DL space, and I hope the communities can evolve. Perhaps the Python ecosystem can improve here over time, but I'm concerned it'll be a low priority and slow going even if it were a priority.