Streaming Operator Inference

This repository contains the code to reproduce the experiments presented in the paper:

Streaming Operator Inference for Model Reduction of Large-Scale Dynamical Systems

This repository contains the 1D viscous Burgers equation experiment and Kuramoto-Sivashinsky equation (KSE) example. The experiments demonstrate the performance of the Streaming-OpInf method for learning reduced-order models from streaming data. The 3D channel data is not available for public release and thus experiments on that dataset are not included here.

Requirements

• You must have Julia installed on your system. You can download it from the official Julia website.

Running the Experiments

To run the experiments, first clone this repository

git clone https://github.com/smallpondtom/StreamingOpInf.git

Prepare the Julia environment

cd StreamingOpInf
julia
julia> ]
(@v1.11) pkg> instantiate 

Exit Julia by typing exit() or pressing Ctrl+D. The command julia will envoke the Julia REPL. The command ] will enter the package manager mode. The command instantiate will install all the required packages specified in the Project.toml file and generate a Manifest.toml file for your dependencies.

Now you are ready to run the experiments. Make sure you are in the root directory of the project, and then execute the appropriate script for your operating system:

• For Windows PowerShell:

.\run_experiments.ps1 -Experiment burgers
.\run_experiments.ps1 -Experiment kse
.\run_experiments.ps1 -Experiment both

• For Windows Command Prompt:

run_experiments.bat burgers
run_experiments.bat kse
run_experiments.bat both

• For Unix-based systems (Linux, macOS):

chmod +x ./run_experiments.sh burgers
chmod +x ./run_experiments.sh kse
chmod +x ./run_experiments.sh both

It is going to take a while to run each experiment so maybe run them overnight.

Dependencies

The experiments rely on the following original Julia packages:

• LiftAndLearn.jl: The main Julia implementation of Operator Inference (and Lift & Learn) methods for learning reduced-order models from data.
• IncrementalSVD.jl: A Julia package for performing incremental singular value decomposition (SVD) on streaming data. Including the key algorithms: Baker's algorithm and SketchySVD.
• ChaosGizmo.jl: A Julia package for computing the Lyapunov exponents and the Kaplan-Yorke dimension. This package is used specifically used for the Kuramoto-Sivashinsky equation experiment to analyze the chaotic behavior of the learned reduced-order models.

Contact

For any questions, please contact Tomoki Koike at tkoike@gatech.edu.