hypersindy 0.0.3

HyperSINDy, a package to discover RDEs from time series data.

HyperSINDy

This repository is the official implementation of HyperSINDy, first introduced in HyperSINDy: Deep Generative Modeling of Nonlinear Stochastic Governing Equations.

Requirements

All the requirements are contained in the environment.yml file\ To install the requirements, run:

conda env create -f environment.yml

Then, activate the conda environment:

conda activate hypersindy

Or, with some conda environment that can run Python 3.9, you can manually install the dependencies:

conda install scipy seaborn tensorboard matplotlib scikit-learn pandas jupyterlab pip
pip3 install pysindy torch==1.12.0 torchvision

Installation

After installing dependencies, you can install hypersindy:

pip3 install hypersindy

Example use case

See example.py as well. `` python3

from hypersindy.library import Library from hypersindy.net import Net from hypersindy.trainer import Trainer from hypersindy.dataset import SyntheticDataset from hypersindy.utils import set_random_seed

set_random_seed(0) device = 2 x_dim = 3 z_dim = 6 data_path = 'x_train.npy'

library = Library(x_dim) net = Net(library, z_dim).to(device) dataset = SyntheticDataset(library, fpath=data_path) trainer = Trainer(net, library, "runs/1", "runs/1.pt", device=device) trainer.train(dataset)

equations results ``

Results can be viewed in tensorboard. tensorboard --logdir="runs"

Paper

To reproduce results from the paper, go to the paper folder and view the README there. cd paper