Pytorch implementation of Modulated Neural ODEs
The code was developed and tested with python3.8 and pytorch 1.13. You can install the relevant packages via
pip install -r requirements.txt
Data is generated on the fly in the folder data. In this folder you can also find the data configuration file: config.yml. In total the code enables to generate 5 different datasets:
- Sinusoidal data
- Predator Prey data
- Bouncing Balls
- Rotating MNIST
- Mocap
Upon generation a folder for the corresponding dataset is created with a train, valid, and test dataset, respectively.
For Mocap data, please download the files reported in the appendix from http://mocap.cs.cmu.edu/subjects.php.
To the train the models run the main.py file with passing the relevant arguments. For example commands for each dataset please see the commands.txt file.
An example commands for sine data with MoNODE model:
python main.py --task sin --model node --solver rk4 --batch_size 10 --T_in 3 --T_inv 10 --ode_latent_dim 4 --modulator_dim 4 --Nepoch 600
All log files, figures and trained model will be stored automatically in a results folder.