Graphnet

Graphnet is a Graph Neural Network (GNN) model for the prediction of the stationnary physical solution of a PDE, given the mesh of the geometry. The framework is based on the PyTorch Geometric library. The network is based on the MeshGraphNets architecture.

Setup

Conda environment

conda env create -f utils/envs/graphnet_no_builds.yml
conda activate graphnet

Download data

The dataset is available. If you have not downloaded yet for Meshnet, in a different folder than the one containing this repository, run:

git clone https://github.com/UgoPelissier/dataset

And follow the instructions in the README.md file. This will create vtu folders inside stokes2, stokes3 and stokes3adapt folders containing the CAD models.

Last step is to move the stokes2, stokes3 and stokes3adapt folders inside the data folder of this graphnet repository. The final structure should look like this:

├── callbacks
├── configs
├── data
└── data
    └── stokes2
    ├── stokes3
    ├── stokes3adapt
        └── msh
            ├── cad_000.msh
            :
            └── cad_500.msh
        └── mesh
            ├── cad_000.mesh
            :
            └── cad_500.mesh
        └── raw
            ├── cad_000.vtu
            :
            └── cad_500.vtu
    ├── datamodule.py
    └── dataset.py
├── model
├── utils
├── __init__.py
├── .gitignore
├── main.py
└── README.md

Train the model

Parameters setting is done in the configs/config.yaml file. Check, and change if needed, the parameters marked with # TODO comments.

Train the model by running:

python main.py fit -c configs/config.yaml

You can get help on the command line arguments by running:

python main.py fit --help

It will create a new folder in the logs/ folder containing the checkpoints of the model and a configuration file containing the parameters used for the training, that you can use later if you want.

Evaluate the model

To evaluate the model training, run:

tensorboard --logdir=logs/

Test the model

To test the model, run:

python main.py test -c configs/config.yaml --ckpt_path $ckpt_path

where $ckpt_path is the path to the checkpoint file located in the logs/version_$version/checkpoints/ folder.

It will create a new folder in the logs/ folder containing the meshes resulting from the predictions of the model (vtk files).

Contact

Ugo Pelissier
<ugo.pelissier@etu.minesparis.psl.eu>