https://github.com/ur-whitelab/fedchem

https://github.com/ur-whitelab/fedchem

Last synced: about 1 month ago
JSON representation

• Host: GitHub
• URL: https://github.com/ur-whitelab/fedchem
• Owner: ur-whitelab
• Created: 2021-12-07T03:06:18.000Z (about 3 years ago)
• Default Branch: main
• Last Pushed: 2022-03-15T05:27:29.000Z (almost 3 years ago)
• Last Synced: 2024-04-15T15:11:12.859Z (9 months ago)
• Language: Python
• Size: 506 KB
• Stars: 3
• Watchers: 3
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# FedChem and FLIT(+)

We provide the script for running FLIT for the proposed benchmark. Our code is developed based on FedML (https://fedml.ai/)

## Requirements

    dgl==0.6.1

    dgllife==0.2.6

    easydict==1.9

    pytorch-geometric==1.7.2

    rdkit=2019.09.3

    pytorch=1.8.1

## Dataset Download (optional)

All dataset will be downloaded with first run or you can download them by

```angular2html

python downloadDataset.py

```

We provide the scaffold splitting results for all datasets and save them at ./data/scaffoldresult/scffoldLabel_xxx.pt

## Usage

You need a gpu to run the code. We log the results with wandb.

1. Train FedAvg for FreeSolv with heterogeneous partatition 0.1 by 

```

python main.py -dataset esol -fedmid avg -part_alpha 0.1

```

2. Train FLIT+ (gamma(tmpFed)=0.5 and lambda(lambdavat)=0.01) for FreeSolv with heterogeneous partatition 0.1 by

```

python main.py -dataset esol -fedmid oursvatFLITPLUS -tmpFed 0.5 -lambdavat 0.01 -part_alpha 0.1

```

## Citation

Cite our paper

```angular2html

@article{zhu2021federated,

  title={Federated Learning of Molecular Properties with Graph Neural Networks in a Heterogeneous Setting},

  author={Zhu, Wei and White, Andrew and Luo, Jiebo},

  journal={Available at SSRN 4002763},

  year={2021}

}

```