In this repository, we implement GEEL: Gap Encoded Edge List in the paper: A Simple and Scalable Representation for Graph Generation.
ALT is built in Python 3.10.0, PyTorch 1.12.1, and PyTorch Geometric 2.2.0 . Use the following commands to install the required python packages.
conda env create --file environment.yaml
The configurations are given in config/trans/
directory. Note that max_len denotes the maximum length of the sequence representation in generation. We set max_len as the maximum number of edges of training and test graphs.
You can train ALT model and generate samples by running:
CUDA_VISIBLE_DEVICES=${gpu_id} bash script/trans/{script_name}.sh
For example,
CUDA_VISIBLE_DEVICES=0 bash script/trans/com_small.sh
Then the generated samples are saved in samples/
directory and the metrics are reported on WANDB.