Code for Data-driven closure discovery

Brief overview

This is the code, written in Python, for implementing several toy cases in "Data driven discovery of closure models" in https://arxiv.org/abs/1803.09318

How to run this code?

To run the code, one would need

• cython
• nolds
• sklearn
• keras
• tensorflow
• sindy package from Jean-Christophe: https://github.com/loiseaujc/SINDy
  • note that initially borrow from this version but later on moved to sklearn implementation of SINDy.
  • readers can also try different version of SINDy but keep in mind regularization coefficient should be well tuned.

After that, one can start to compiled the code using Cython, as a purpose to acclerating our Python code.

• cd to /src folder, run make

To run different cases, one should first cd to /src folder, and then doing the folloing.

To run simple 3D linear system case

• python cython_main.py 3d_linear sindy 1 tdf 1

To run 2D VDP case

• python cython_main.py 2d_vdp sindy 0 tdf 3

To run Lorenz case

• non-chaotic case:
  • python cython_main.py lorenz-equ ann_cross_time 1 18
• chaotic case:
  • python cython_main.py lorenz-chaos ann_cross_time 1 18

How to check final result?

To check the final result, go into the folder of the /data/ and check corresponding /image subfolders

How long does it take on a normal laptop?

For 3D linear and 2D VDP case, it is really fast within 2-5 mins.

For Lorenz cases, since the number of data points is a bit large and we have a posteriori evaluations, the computational time is a bit long. For non-chaotic cases it took around 30 mins and for chaotic case it took 220 mins.

How to play around with it and tune different parameters?

Check out /src/config.pyx