MyPaperCode

This project is organized in a HAFusion-style runtime layout while keeping the HRE model code unchanged. The New York experiment now uses HAFusion-style structural inputs: POI similarity, mobility adjacency, land-use similarity, and the original visual embedding.

Structure

• data_NewYork/: New York dataset files used by the current project
• tasks_NewYork/: downstream task scripts for the New York dataset
• project_config.py: dataset and task registry
• parse_args.py: unified CLI parser
• HRE_Module_Train.py: training entry with dataset/task dispatch
• main.py: simple launcher

Quick Start

Train for a single downstream task:

python main.py --city NewYork --task crime
python main.py --city NewYork --task checkin
python main.py --city NewYork --task servicecall

Run all downstream tasks after training, while selecting the best checkpoint with a specific task:

python main.py --city NewYork --task all --selection_task crime

Useful runtime parameters:

python main.py --city NY --task crime --epochs 500 --learning_rate 0.0002 --dropout 0.2

Outputs are saved under outputs/<city>/<task>/ by default.