search

asteroid-team / torch-audiomentations

Fast audio data augmentation in PyTorch. Inspired by audiomentations. Useful for deep learning.
MIT License
969 stars 88 forks source link

Make a shared requirements.txt for tests & build #105

Closed KentoNishi closed 3 years ago

KentoNishi commented 3 years ago

Currently, updating requirements needs to be done in two files separately. Why not just make a shared requirements.txt file for the package dependencies and install them both in CI tests? I moved the setup.py requirements list into requirements.txt which is also referenced when initializing the environment in actions.

iver56 commented 3 years ago

I appreciate the idea of specifying the dependencies in a DRYer way, but I am hesitant to accept these changes. Here are my thoughts:

1) I want to keep using conda for the dev environment, because I've never been able to install a GPU-enabled pytorch without it. I feel like conda works well, so I haven't had the motivation to "downgrade" to using only pip 2) The package dependencies specified in setup.py must allow a wide range of dependency versions while the CI and dev environment should use pinned packages

If we should make the dependencies list DRYer, I feel like the CI should use conda and environment.yml (instead of pip and a test_requirements.txt) to set up its environment. Then we'll have two lists instead of three. But let's not boil it down to one.

KentoNishi commented 3 years ago

fair points, ill close the issue then 👍

also I've never been able to install GPU torch through pip either so we're in the same boat 😂

iver56 commented 3 years ago

Ok :) I have added an issue here: https://github.com/asteroid-team/torch-audiomentations/issues/106