Code for reproducing experiments from Learning continuous-time PDEs from sparse data with graph neural networks.
Data can be downloaded here.
You will need to install graphpdes before running the scripts.
The data comes in the following format:
- t (time grid) ndarray with shape (simulations, timepoints)
- x (node positions) ndarray with shape (simulations, nodes, grid dimension)
- u (field) ndarray with shape (simulations, timepoints, nodes, field dimension)
- bcs_dicts (only for Dirichlet BCs) - dictionary {'bc_name': [[bc_node_indices], [field_indices]], etc}. Used to zero the predicted time derivative of u at nodes in bc_node_indices and field dimensions in field_indices.