This repository contains the source code accompanying the paper
CHAROT: Robustly controlling chaotic PDEs with partial observations
Accepted at the ICLR 2024 Workshop on AI4Differential Equations in Science.
First install the requirements by running pip install -r requirements.txt
. (If you are running on MacOS with an M1 chip, you need to install torchrl 0.2.0 from the GitHub repo.)
To run a training script, simply run python tqc.py
.
All relevant hyperparameters may be changed in 'config.yaml'. In particular, to select which augmentation of TQC should be run, change the network.architecture parameter to one of 'base', 'attention' (= CHAROT) or 'lstm'.