Nested Fourier Neural Operater for Darcy Flow

This example demonstrates how to set up a data-driven model for a 2D Darcy flow using the Nested Fourier Neural Operator (FNO) architecture inside of Modulus. Training progress can be tracked through MLFlow. This case is parallelised to run in multi-GPU settings.

Start with generating the dataset for training:

            

            
python generate_nested_darcy.py
        

which will create the folder ./data with out_of_sample.npy, training_data.npy, validation_data.npy.

To train the model on a single GPU, run

            

            
python train_nested_darcy.py +model=ref0
python train_nested_darcy.py +model=ref1
        

For training a model on two GPUs, run

            

            
mpirun -n 2 python train_nested_darcy.py +model=ref0
mpirun -n 2 python train_nested_darcy.py +model=ref1
        

To evaluate the model use:

            

            
python evaluate_nested_darcy.py
        

Progress can be monitored using MLFlow. Open a new terminal and navigate to the training directory, then run:

            

            
mlflow ui -p 2458
        

View progress in a browser at http://127.0.0.1:2458

If training on a remote machine, set up a ssh tunnel to the server with LocalForward 8080 your_remote_machine_addr:8080. ssh to the server via the specified port, in this case 8080, navigate to the training directory and launch mlflow server

            

            
mlflow server --host 0.0.0.0 --port 8080
        

On your local machine, open a browser and connect to localhost:8080.

• Real-time high-resolution CO2 geological storage prediction using nested Fourier neural operators

• Fourier Neural Operator for Parametric Partial Differential Equations