OperatorFunctionEncoder

This is the official repo for "Basis to Basis Operator Learning via Function Encoders". See the project page for more information.

Getting Started

First, install torch using this website's command: https://pytorch.org/get-started/locally/

Then, install all packages using pip:

pip install FunctionEncoder==0.0.4 numpy matplotlib tqdm scipy tensorboard

Download data using the following commands:

python src/Datasets/DarcyDataset.py
python download_data.py

Working Directory

All commands are run from OperatorFunctionEncoder, the base working directory. Do not change into src/ or run_scripts/ or plotting_scripts/, this will likely break things.

Running the code

To run the code for one example, run the following command:

python test.py (args)

Then are numerous arguments to select different algs and datasets.

Alternatively, use the following to run all experiments:

chmod +x ./run_scripts/run_ablation_n_basis.sh # makes it executable
chmod +x ./run_scripts/run_ablation_n_sensor.sh 
chmod +x ./run_scripts/run_ablation_unfreeze.sh 
chmod +x ./run_scripts/run_all.sh 
chmod +x ./run_scripts/run_experiment.sh 
./run_scripts/run_all.sh

You will likely have to change the arguments at the top of the file to 1 Gpu.

Note this will take a long time to run.

Experimental Results

Training curves for the experiments in the paper are available here