ALDD

Adaptive Local Domain Decomposition for neural network learning of large-scale partial differential equations

Introduction

This repository contains the code for the paper "Energy-based feature extraction with adaptive local domain decomposition for prediction of transient and turbulence flow with operator regression models".

Quick Start

To configure the environment, run the following command:

pip install -r requirements.txt

Training and prediction on ALDD, naive Data Scoping (DS), and full data (one-shot) can be done by running the three separate python scripts with appropriate configurations. One example experiment doing prediction on the John Hopkins Turbulence Database (JHTDB) with ALDD, with pre-trained PCA encoder, K-Means clustering, and Fourier Neural Operator (FNO) model, can be done by running the following command:

python3 run_ALDS.py --dataset=jhtdb --encoder=pca --classifier=kmeans --model=fno --exp_name=fno_jhtdb_aldd --mode=pred --exp_config configs/exp_config/fno_jhtdb.yaml --train_config configs/train_config/fno.yaml

After executing the command, the prediction results, figures as well as raw data will be saved in the folder ./logs/.