IFH: a Diffusion Framework for Flexible Design of Graph Generative Models

This is the code which was used to run the experiments for the paper "IFH: a Diffusion Framework for Flexible Design of Graph Generative Models".

Requirements

The code was tested with Python 3.11.7. The requirements can be installed by executing

• Download anaconda/miniconda if needed
• Create a new environment through the given environment files with the following command:

  conda env create -f <env_file>.yml

  where <env_file> is the name of the environment file to use. It is possible to install dependencies for CPU with environment_cpu.yml or for GPU with environment_cuda.yml.
• Install this package with the following command:

  pip install -e .

  which will compile required cython code.
• Navigate the directory ./src/metrics/utils/orca and compile the orca executable with the following command:

  g++ -O2 -std=c++11 -o orca orca.cpp

Experiments

For CUDA>=10.2, to run any experiments in a reproducible way, it is necessary to set the environment variable:

    export CUBLAS_WORKSPACE_CONFIG=:4096:8

or in Windows, for cmd:

    set CUBLAS_WORKSPACE_CONFIG=:4096:8

for PowerShell:

    $env:CUBLAS_WORKSPACE_CONFIG=":4096:8"

and to remove:

    Remove-Item Env:\CUBLAS_WORKSPACE_CONFIG

Hyperparameters searches can be performed with the commands:

    python main_multirun.py +experiment/seq_degree=<degree> +experiment/task=<dataset> seed=<seed> +modsel=<hparams>

where <degree> is the sequentiality degree, found in experiment/seq_degree; <dataset> contains all the dataset task information; <seed> is the seed to use for the experiment; <hparams> contains the set of hyperparameters to search on, i.e., "bayesian_no_reins" for 1-node sequential models, "bayesian_15" for block generation, and "bayesian_no_halt_reins_15" for one-shot models.

All the configurations of the experiments can be found at ./src/configs/compact. These files also contain the hyperparameters found in the search procedure. The experiments can be run with different seeds with the following command:

    python main.py +compact/<dataset>=<experiment> seed=<seed>

where <dataset> is the name of the dataset, <experiment> is the name of the experiment and <seed> is the seed to use for the experiment. For example, to run the experiment "exp4_one_zinc" on the dataset "zinc" with the seed 2, the following command can be used:

    python main.py +compact/zinc=exp4_one_zinc seed=2

Datasets

Datasets will be downloaded automatically to a new directory ./datasets when running the experiments. Current available datasets are QM9, ZINC250K, community, ego-small, enzymes, and ego. The datasets are then stored in the PyTorch Geometric format.

Checkpoints and logging

Checkpoints are saved in a new directory ./checkpoints, and logging is done through WandB, which requires a free account to be used.