SINDy Autoencoder - PyTorch

PyTorch implementation of the SINDy Autoencoder from the paper "Data-driven discovery of coordinates and governing equations" by Champion et al.

The original implementation is in TensorFlow and is located at: https://github.com/kpchamp/SindyAutoencoders
The TensorFlow implementation was used as a reference and, for some components, code was directly copied over. In the latter case, each file with copied code has a reference to which file in the original repository it was taken from.

Datasets

Currently, the model supports only using a Lorenz system dataset. To create it, run "python3 create_lorenz.py"

To run the model:

python3 main.py

To run the model from a previous checkpoint:

python3 main.py --load_cp 1

To run the experiment (automatically loads the checkpoint):

python3 experiments.py

cmd_line.py

cmd_line.py contains arguments that can be adjusted for the model (ie, learning rates and loss lambdas)
The easiest way to do experiments is to adjust parameters in cmd_line.py and then run python3 main.py.