Convention for specifying dependencies

[jsh9]

The convention for specifying dependencies is usually a requirements.txt file like this:

pytorch
torchvision
torchaudio
pytorch-cuda==11.8
pandas
...

And then let people install them by pip install -r requirements.txt.

Conda may not be necessary, pip should be sufficient.

Also, it seems torchvision and torchaudio are not needed in the algorithm? There's no need to specify unused dependencies in the requirements.txt file.

[yzshi5]

Hi Jian, I agree with that a requirements.txt may be necessary and more formal. Since this work is a pilot study, and has several follow ups. I will be more careful in further code release.

Usually, build a separate conda environment is a convenient way for testing a new model, especially for people without a good background in computer system. For the installation of unused torchvision and torchaudio, I just copied the command line for installing pytorch from the official website.

[jsh9]

Usually, build a separate conda environment is a convenient way for testing a new model, especially for people without a good background in computer system.

Yeah, people can just do pip install -r requirements.txt after they create a fresh conda environment.