Diffusion Twigs with Loop Guidance for Conditional Graph Generation

Giangiacomo Mercatali* | Yogesh verma* | Andre Freitas | Vikas Garg

This repository includes the supporting code for:

Giangiacomo Mercatali*, Yogesh Verma*, Andre Freitas, Vikas Garg. Diffusion Twigs with Loop Guidance for Conditional Graph Generation. In Advances in Neural Information Processing Systems 38, 2024.

Twigs

Environment installation

Install packages in env.yml. Tested on pytorch 1.13.1 py3.8_0

QM9 Dataset

Download preprocessed data (by Huang et al 2023) found at this link into the data folder.

Model Arguments

Use the option --config.model.name=cond_DGT_concat to run Jodo from Huang et al. 2023.

Train Twigs: 2 properties

Pairs: (Cv,Mu), (Gap,Mu), (alpha,Mu).

CUDA_VISIBLE_DEVICES=0 python main.py --config configs/vpsde_qm9_cond_multi_twigs.py --config.model.name=cond_DGT_twigs --config.training.n_iters=3000000 --mode train --config.nprops=2 --config.model.cond_ch=2 --workdir exp_cond_multi/vpsde_qm9_cond_twigs_Cv_mu --config.cond_property1 Cv --config.cond_property2 mu
CUDA_VISIBLE_DEVICES=0 python main.py --config configs/vpsde_qm9_cond_multi_twigs.py --config.model.name=cond_DGT_twigs --config.training.n_iters=3000000 --mode train --config.nprops=2 --config.model.cond_ch=2 --workdir exp_cond_multi/vpsde_qm9_cond_twigs_gap_mu --config.cond_property1 gap --config.cond_property2 mu --config.training.snapshot_freq=100000
CUDA_VISIBLE_DEVICES=0 python main.py --config configs/vpsde_qm9_cond_multi_twigs.py --config.model.name=cond_DGT_twigs --config.training.n_iters=3000000 --mode train --config.nprops=2 --config.model.cond_ch=2 --workdir exp_cond_multi/vpsde_qm9_cond_twigs_alpha_mu --config.cond_property1 alpha --config.cond_property2 mu --config.training.snapshot_freq=100000

Sample Twigs: 2 properties

ckpt=100
CUDA_VISIBLE_DEVICES=0 python main.py --config configs/vpsde_qm9_cond_multi_twigs.py --config.model.name=cond_DGT_twigs --mode eval --config.nprops=2 --config.model.cond_ch=2 --workdir exp_cond_multi/vpsde_qm9_cond_twigs_Cv_mu --config.cond_property1 Cv --config.cond_property2 mu --config.eval.save_graph=True --config.eval.ckpts=$ckpt
CUDA_VISIBLE_DEVICES=0 python main.py --config configs/vpsde_qm9_cond_multi_twigs.py --config.model.name=cond_DGT_twigs --mode eval --config.nprops=2 --config.model.cond_ch=2 --workdir exp_cond_multi/vpsde_qm9_cond_twigs_gap_mu --config.cond_property1 gap --config.cond_property2 mu --config.eval.save_graph=True --config.eval.ckpts=$ckpt
CUDA_VISIBLE_DEVICES=0 python main.py --config configs/vpsde_qm9_cond_multi_twigs.py --config.model.name=cond_DGT_twigs --mode eval --config.nprops=2 --config.model.cond_ch=2 --workdir exp_cond_multi/vpsde_qm9_cond_twigs_alpha_mu --config.cond_property1 alpha --config.cond_property2 mu --config.eval.save_graph=True --config.eval.ckpts=$ckpt

Train Twigs: 3 properties

properties: alpha,mu,gap

CUDA_VISIBLE_DEVICES=0 python main.py --config configs/vpsde_qm9_cond_multi_twigs.py --config.model.name=cond_DGT_twigs --config.training.n_iters=3000000 --mode train --config.nprops=3 --config.model.cond_ch=3 --workdir exp_cond_multi/vpsde_qm9_cond_twigs_alpha_mu_gap --config.cond_property1 alpha --config.cond_property2 mu --config.cond_property3 gap --config.training.snapshot_freq=100000

Sample Twigs: 3 properties

CUDA_VISIBLE_DEVICES=0 python main.py --config configs/vpsde_qm9_cond_multi_twigs.py --config.model.name=cond_DGT_twigs --mode eval --config.nprops=3 --config.model.cond_ch=3 --workdir exp_cond_multi/vpsde_qm9_cond_twigs_alpha_mu_gap --config.cond_property1 alpha --config.cond_property2 mu --config.cond_property3 gap --config.eval.save_graph=True --config.eval.ckpts=$ckpt

Citation

If you find this repository useful in your research, please consider citing the following paper:

@inproceedings{
diffusiontwigs,
title={Diffusion Twigs with Loop Guidance for Conditional Graph Generation},
author= {Mercatali, Giangiacomo and Verma, Yogesh and Freitas, Andre and Garg, Vikas},
booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
year={2024},
url={https://openreview.net/forum?id=fvOCJAAYLx}
}