4 ms·

Abstractions always leak somehow. For example at work we had specific OS related python issues (dependent libraries to vizualize and save plotly graphics) despi

• PaulHoule 4y ago
  After a year of using it heavily I came to see conda as more a problem than a solution. For one thing, compressing the files with bzip made installations of large packages really slow. Then there was the time I found that things were breaking all the time because the prima donna data scientists were still doing pip install --user which was breaking all of their conda environments. (Granted it's a problem that affects normal virtualenvs but it was one of many of the real problems that conda left unaddressed.) The one thing I liked about conda was that it could install optimized versions of libraries like numpy but really the right answer for that is that you should be able to tell the package solver that "numpy-mkl" satisfies "numpy". It was definitely a blast to make conda packages for different versions of the NVIDIA drivers (100% userspace libs!) so I could run models written in various versions of Tensorflow without having to fight with NVIDIA's installers... Even if it took way too long to decompress the bzip files. It was a shame that conda wasn't given permission to do the same because it would have been a real selling point.