Differential-Privacy-and-LQR

ACC 2023 submission. link

Multiple differential Private forecasters for LQR.

Prerequisites

python version: 3.9.12

package	version
numpy	1.23.1
pandas	1.4.4
PyYAML	6.0
matplotlib	3.5.3
tsai	0.3.1
torch	1.11.0
scikit-learn	1.1.2
seaborn	0.12.0
cvxpy	1.2.1

For more details, see environment.yaml.

Getting Started

Pls run all the source file in the root directory :)

Experiments

• Go to the root directory of the repo.

ARIMA

• To run experiments: arima_syn.py

• To plot: arima_merge_plot.py

Uber

• Pre-process uber data from the "uber data" directory. Please download the data here, and execute: uber_preprocessing.py

• To train timeseries forecast models: uber_train_forecaster.py

• To run experiments: uber_syn.py

• To plot: uber_merge_plot.py

Other functions

• Src code of input-driven LQR controllers and ARIMA time series: system_dynamics.py

• Dimension of action space, state space, and other parameters: parameters.yml

• Functions to plot: plot_utilities.py

• Alternative Convex Search algorithm: ACS.py

• Class for timeseries sources: src_class.py

Contact

Po-han Li - pohanli@utexas.edu