SAND

The official PyTorch implementation of "Learning to Simulate Daily Activities via Modeling Dynamic Human Needs" (WWW'23).

The code is tested under a Linux desktop with torch 1.7 and Python 3.7.10.

Installation

Environment

• Tested OS: Linux
• Python >= 3.7
• PyTorch == 1.7.1

Dependencies

1. Install PyTorch 1.7.1 with the correct CUDA version.
2. Use the pip install -r requirements. txt command to install all of the Python modules and packages used in this project.

Model Training

Use the following command to train SAND on the Foursquare dataset:

cd SAND;

python app.py --dataset 'Foursquare' --mode 'train'

or on the Mobile Operator dataset:

python app.py --dataset 'Mobile' --mode 'train'

or on the Synthetic Operator dataset:

python app.py --dataset 'Synthetic' --mode 'train'

The trained models are saved in model/TIME/.

Simulation

Use the following command to generate activity data on the Foursquare dataset:

cd SAND;

python app.py --dataset 'Foursquare' --mode 'generate' --generate_final_path your_path

Please specify your own path by the command-line argument generate_final_path for saving the generated data. Then the generated activity data will be in your_path/gen_data.json.

More Related Works

• Activity Trajectory Generation via Modeling Spatiotemporal Dynamics (KDD '22)

Note

The implemention is based on NJSDE.

If you found this library useful in your research, please consider citing:

@inproceedings{yuan2023learning,
  title={Learning to Simulate Daily Activities via Modeling Dynamic Human Needs},
  author={Yuan, Yuan and Wang, Huandong and Ding, Jingtao and Jin, Depeng and Li, Yong},
  booktitle={Proceedings of the ACM Web Conference 2023},
  pages={906--916},
  year={2023}
}