Skip to content
/ LENS Public

Codebase for "Identifying nonequilibrium degrees of freedom in high-dimensional stochastic systems" by Catherine Ji*, Ravin Raj*, Ben Eysenbach, Gautam Reddy.

Notifications You must be signed in to change notification settings

cj7280/LENS

Repository files navigation

Setting up the evironment

To set up environment, run the following commands in terminal on the login node:

CONDA_OVERRIDE_CUDA="12.0" conda create --name env_jrl python=3.10 jax==0.4.23 "jaxlib==0.4.23=cuda120*" flax==0.7.4 numpy==1.26.4 scipy==1.12 matplotlib imageio pytorch -c conda-forge -c nvidia 

pip install tyro wandb==0.17.9 wandb_osh==1.2.2  brax==0.10.1 mediapy==1.2.2 scipy==1.12.0 pyyaml h5py torch

or run the following which directly creates the environment from the yaml:

conda env create -f environment.yml

A todo is to clean up the the environment and get rid of brax.

File organization

All main training files live in src folder and training.py file.

The dataset_creation folder contains code to generate harmonically-coupled beads, render these beads, simulate 2D Complex Ginzburg-Landau dynamics, and simulate a driven bead in a periodic potential.

About

Codebase for "Identifying nonequilibrium degrees of freedom in high-dimensional stochastic systems" by Catherine Ji*, Ravin Raj*, Ben Eysenbach, Gautam Reddy.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published