Dynamic Public Resource Allocation based on Human Mobility Prediction

January 25, 2020 ยท View on GitHub

Paper

If you find our code or dataset useful for your research, please cite our paper:

Sijie Ruan, Jie Bao, Yuxuan Liang, Ruiyuan Li, Tianfu He, Chuishi Meng, Yanhua Li, Yingcai Wu and Yu Zheng. "Dynamic Public Resource Allocation based on Human Mobility Prediction.", ACM IMWUT/UbiComp 2020.

Requirements

Python 3.6

  • numpy==1.14.5
  • networkx==2.2
  • shapely==1.6.4
  • pickle

Dataset

We organize our dataset into two archives, i.e., MALMCS_data.zip and PREDICTION_data.zip

  1. MALMCS_data.zip

  • frames_20180101_20181101_24.npy: this is the hourly crowd flows data in Beijing Happy Valley from 01/01/2018 to 01/11/2018 scraped from the Tencent Heat Map. The last month is used for evaluation, and previous months are used for training & validation.

  • pred_all_stresnet_mf4_masked.pkl: this is the predicted results from the prediction model for evaluation acceleration purpose. In the paper, those results are obtained by training MF-STN.

  1. PREDICTION_data.zip

This archive provides some external factors for crowd flow prediction, which can be used to train the crowd flow prediction model together with frames_20180101_20181101_24.npy. This dataset is also a data source for UrbanFM.

  • holiday features: external/holiday_20180101_20181101_24.npy
  • meteorology features: external/mete_cy_20180101_20181101_24.npy
  • ticket price features: external/price_20180101_20181101_24.npy
  • time of day features: external/tod_20180101_20181101_24.npy

Usage

Tunable Parameters

  • Service radius radius
  • Energy limitation cost_limit
  • Number of agents k
python evaluate.py

License

The code and data are released under the MIT License.