Liyue-Chen / STCFPContext

STCFP Context

Project

This repository provides the data and code for reproducing the experiment results of Benchmarking Context Factor Generalizability in Spatiotemporal Crowd Flow Prediciton.

Data

We provide crowd flow and corresponding POI data in the dataset directory. Move them into the UCTB package data directory to make the dataloader could find the data files. Weather data is stored in 'External' key of the *.pkl files.

Google Drive

Environment

Urban Computing Tool Box (UCTB, https://github.com/uctb/UCTB) is a package providing spatial-temporal prediction models. This project is developed based on the UCTB toolkit, so firstly install UCTB by the following command.

python3 build_install.py

Experiments require specific dependencies.

• tensorflow-gpu ==1.13.1
• keras ==2.2.4
• cuda toolkit==10.0

Exploring the generalizability of different modeling techniques

To analyze context modeling techniques, run the following experiments.

Experiments based on STMGCN

In STCFP_External\Experiments\STMGCN directory, run the following commands.

python3 Runner_techniques_analysis_30_STMGCN.py

python3 Runner_techniques_analysis_60_STMGCN.py

python3 Runner_techniques_analysis_120_STMGCN.py

Experiments based on STMeta

In STCFP_External\Experiments\STMeta directory, run the following commands.

python3 Runner_features_analysis_30_STMeta.py

python3 Runner_features_analysis_60_STMeta.py

python3 Runner_features_analysis_120_STMeta.py

Exploring the generalizability of context features

To exploring the generalizability of context features, run the following experiments.

Experiments based on XGBoost

In STCFP_External\Experiments\XGBoost directory, run the following commands.

python3 Runner_features_analysis_XGBoost_30.py

python3 Runner_features_analysis_XGBoost_60.py

python3 Runner_features_analysis_XGBoost_120.py

Experiments based on STMGCN

In STCFP_External\Experiments\ST_MGCN directory, run the following commands.

python3 Runner_features_analysis_30_STMGCN.py

python3 Runner_features_analysis_60_STMGCN.py

python3 Runner_features_analysis_120_STMGCN.py

Experiments based on STMeta

In STCFP_External\Experiments\STMeta directory, run the following commands.

python3 Runner_features_analysis_30_STMeta.py

python3 Runner_features_analysis_60_STMeta.py

python3 Runner_features_analysis_120_STMeta.py