In this project we attempt to use a modified Fourier Neural Operator to Resolve Shock Conditions in the Euler System of PDEs. Additional data files/models/etc. can be found on our Google Drive
The reporistory is organized as follows:
train_model.py
: Script for training the models. Draws configurations and models fromconfigs/config_train.py
eval_models.py
: Evaluates a single model according toconfigs/config_eval.py
eval_all_upscales.py
: Evaluates all of the models listed inconfigs/config_all_upscales.py
and outputs a.mat
file toeval/
plot_epochs.py
: Plot the train/test MSE over the epochs, from the data stored inpred/
plot_upscales.py
: Plots the data from the models/losses inconfigs/config_all_upscales.py
for increasing reolutions, as specified in the config fileEulerDataGen.m
: Generate boundary conditions/solutions to the Euler Fluid equations (system of PDEs)
configs/
: Config files for training/evaluation of models. Must set before running most of the above scriptsdata/
: Train/test data for training and evaluating the models. Can be generated withEulerDataGen.m
or downloaded form the Google Driveeval/
: Model outputs and losses/MSEs on test dataplots/
: Output directory for plotting scriptspred/
: Predictions on test set generated immediately after training, helpful for visualizing outputs usingEulerDataGen.m
report/
: PDF and slides for the project deliverablesscripts/
: Helper files for training and testing models. Includes the modified FNO architecture, basline FNO, U-Net, loss functions, file readers, and Adam optimizer