Installation Guide

This guide will help you install and set up the COSIE package.

Installation

For convenience, we recommend creating and activating a dedicated conda environment before installing COSIE. If you haven’t installed conda yet, we suggest using Miniconda, a lightweight distribution of conda.

conda create -n cosie_env python=3.9.19
conda activate cosie_env

The COSIE package can be downloaded by:

git clone https://github.com/weilicode/cosie.git
cd cosie

The cosie_env environment can be used in Jupyter Notebook by:

pip install ipykernel
python -m ipykernel install --user --name=cosie_env

Dependencies

COSIE is a graph deep learning model built upon torch-2.4.0 and torch_geometric-2.5.3.

Using GPU acceleration can significantly speed up the training process. If you plan to use a GPU, please make sure that PyTorch and PyTorch Geometric are installed with versions that are compatible with your local CUDA version. For example, if you are using CUDA 12.1, you can install the required packages as follows:

pip install torch==2.4.0 torchvision==0.19.0 torchaudio==2.4.0 --index-url https://download.pytorch.org/whl/cu121
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.4.0+cu121.html
pip install torch_geometric==2.5.3

All other required packages are listed in requirements.txt. You can install them by running:

pip install -r requirements.txt