search

tristandeleu / pytorch-meta

A collection of extensions and data-loaders for few-shot learning & meta-learning in PyTorch
https://tristandeleu.github.io/pytorch-meta/
MIT License
1.98k stars 256 forks source link

Allow installing torchmeta with conda #63

Closed brando90 closed 4 years ago

brando90 commented 4 years ago

doesn't download anything...use pip

brando90 commented 4 years ago

https://stackoverflow.com/questions/41060382/using-pip-to-install-packages-to-anaconda-environment

brando90 commented 4 years ago

allowing forge didn't help:

https://stackoverflow.com/questions/48493505/packagesnotfounderror-the-following-packages-are-not-available-from-current-cha

brando90 commented 4 years ago

is it really safe to only allow to install it via pip? Why can't we also install it via conda?

brando90 commented 4 years ago

https://stackoverflow.com/questions/56134588/is-that-a-bad-idea-to-use-conda-and-pip-install-on-the-same-environment

brando90 commented 4 years ago

maybe this will help:

https://www.colorado.edu/earthlab/2019/01/03/publishing-your-python-code-pip-and-conda-tips-and-best-practices

tristandeleu commented 4 years ago

As far as I understand, and according to your last link, there is no advantage to install the package with conda over pip (since Torchmeta is a "pure python" package). I think by default pip is included when you create a new conda environment, so you can use pip install torchmeta in your conda environment. Do you have any specific requirement for using conda over pip here?

brando90 commented 4 years ago

As far as I understand, and according to your last link, there is no advantage to install the package with conda over pip (since Torchmeta is a "pure python" package). I think by default pip is included when you create a new conda environment, so you can use pip install torchmeta in your conda environment. Do you have any specific requirement for using conda over pip here?

Conda works better with jupyter as far as I know and data science in general.

In the end for me the point is consistency and better management of the packages. Having multiple packages being installed by different tools is usually bad practice (e.g. if I start having issues with a specific package but can't figure out which one I used to install it, it becomes much harder to fix it since any action I take from there might be with the wrong env manager that might cause further unexpected problems).

I'm sure there are more reasons but that's the first one that comes up in my head.

Everyone uses conda afaik for data science so it's good to allow us to install it with conda.

tristandeleu commented 4 years ago

If you are concerned about the interoperability between conda and pip, according to this answer on Stackoverflow this has been improved in newer versions of conda.

Unfortunately I am not familiar with deployment on conda, and I am not comfortable with maintaining two separate releases. Since installing Torchmeta with pip works well in conda (just make sure the pip command you use is the one from your conda environment), I recommend using pip to install the package.

brando90 commented 4 years ago

SO: https://stackoverflow.com/questions/64050737/how-does-one-install-torchmeta-using-conda