DGL-clustering

An example for DGL cluster/subgraph manipulation

1. Download and install the CUDA 11.6 environment, add the environment to path

1.1. Download the cuda 11.6 library, you can refer to https://developer.nvidia.com/cuda-toolkit-archive, my system is x86_64, Ubuntu 18.04, refer to "runfile (local)" for download link

mkdir cuda_install
cd cuda_install
wget https://developer.download.nvidia.com/compute/cuda/11.6.2/local_installers/cuda_11.6.2_510.47.03_linux.run

1.2 Install the cuda 11.6 library to /opt/cuda-11.6 path

sudo sh cuda_11.6.2_510.47.03_linux.run --silent --toolkit --installpath=/opt/cuda-11.6 --override

1.3 Set up environment variables:

To switch between CUDA versions, you'll need to update your environment variables. You can create a shell script or an alias to easily switch between versions.

Create a script (e.g. “switch_cuda.sh”):

#!/bin/bash
export CUDA_HOME=/opt/cuda-$1
export PATH=$CUDA_HOME/bin:$PATH
export LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH

Make the script executable:

chmod +x switch_cuda.sh

Now, you can easily switch between CUDA versions by running:

source switch_cuda.sh <version>

In this example, you need to activate CUDA 11.6:

source switch_cuda.sh 11.6

1.4 Verify installation:

To check if the correct version is active, run:

nvcc --version

This should display the currently active CUDA version.

2. Steps for building an environment for both torch+tensorflow+dgl+pyg

2.1. Create a new conda environment

conda create --prefix ${HOME}/.conda/envs/torch_tf_pyg_dgl python=3.8

2.2. Activate the environment

conda activate torch_tf_pyg_dgl 

2.3. Install pytorch on CUDA 11.6 (please have CUDA 11.6 install)

pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu116

2.4 Install the pyg library according to https://github.com/pyg-team/pytorch_geometric

pip install torch-scatter -f https://data.pyg.org/whl/torch-1.12.1+cu116.html 
pip install torch-sparse -f https://data.pyg.org/whl/torch-1.12.1+cu116.html 
pip install torch-cluster -f https://data.pyg.org/whl/torch-1.12.1+cu116.html
pip install torch-spline-conv -f https://data.pyg.org/whl/torch-1.12.1+cu116.html
pip install torch-geometric

2.5 Install the dgl and tensorflow:

conda install -c "dglteam/label/cu116" dgl
pip install tensorflow==2.12.*

3. Run the example:

python GNN_partition_dgl.py

The code will create a test folder to contain the partitioned graph output.

4. Run the GCoD code:

Reference: https://github.com/GATECH-EIC/GCoD Some minor package compatibility issues are fixed.

cd GCoD
bash GCN_cora.sh

5. Run the pyg example:

2-layer GCN on Cora, Citeseer and PubMed datasets Reference: https://github.com/pyg-team/pytorch_geometric/blob/master/examples/gcn.py