https://github.com/locuslab/cfd-gcn

https://github.com/locuslab/cfd-gcn

Last synced: about 2 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/locuslab/cfd-gcn
• Owner: locuslab
• Created: 2020-07-07T21:53:13.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2023-03-28T13:47:50.000Z (almost 2 years ago)
• Last Synced: 2023-05-16T11:15:36.006Z (over 1 year ago)
• Language: Python
• Size: 2.2 MB
• Stars: 89
• Watchers: 9
• Forks: 30
• Open Issues: 5
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Combining Differentiable PDE Solvers and Graph Neural Networks for Fluid Flow Prediction

This is the repository for the ICML 2020 paper [Combining Differentiable PDE Solvers and Graph Neural Networks for Fluid Flow Prediction](https://arxiv.org/abs/2007.04439) by Filipe de Avila Belbute-Peres, Thomas D. Economon and J. Zico Kolter. If you like our work and want to cite it, you can use

```

@inproceedings{belbute_peres_cfdgcn_2020,

  author = {Belbute-Peres, Filipe de Avila and Economon, Thomas D. and Kolter, J. Zico},

  title = {Combining {Differentiable} {PDE} {Solvers} and {Graph} {Neural} {Networks} for {Fluid} {Flow} {Prediction}},

  booktitle = {International Conference on Machine Learning (ICML)},

  year = {2020}

}

```

The easiest way to run the experiments is to use the Dockerfile contained in this repository. To run straight from source, you can follow the procedures performed in the Dockerfile, though compatibility between dependencies might be harder to achieve.

## Dockerfile

To build the docker image, run

```

docker build --rm -t cfd-gcn .

```

This will create an image with a working version of this repository, with all dependencies installed.

To access the repository in the container you can then run

```

docker run --gpus '"device=0"' -v $PWD/logs/:/cfd-gcn/logs -u $(id -u):$(id -g) --ipc=host -it --rm cfd-gcn

```

This uses GPU 0 (`device=0`) and the current user (`-u $(id -u):$(id -g)`), which can be changed. 

Logs will be saved to the `logs` directory in the current directory (`$PWD/logs`).

To run the experiments, you can either simply run the script

```

sh run.sh

```

or copy the commands inside that script to run a particular experiment.

*Note:* Warning messages of the type `Read -1, expected XXXX, errno = 1` may show up, but are not harmful to execution. 

These can be disabled or piped to `/dev/null` if desired.