MLSimp

This is the implementation code for "Quantifying Point Contributions: A Lightweight Framework for Efficient and Effective Query-Driven Trajectory Simplification".

Environment Requirements

• Python >= 3.10
• Recommended: Latest versions of PyTorch and PyTorch Geometric
• Other dependencies: tqdm, path, rtree

Dataset

• Download the GeoLife dataset from here and extract it to the ./TrajData folder.
• Preprocess the database using python Utils.preprocessing_trajs.py.
• Generate training and testing sets using python Utils.dataset.py.

T-BERT Pretraining

• Run BertPretrain.py to perform pretraining.
• Every 100 epochs, the model will be saved in the ./ModelSave/{dataset}/pretrain folder.

MLSimp Training

• Run MLTrain.py to train GNN-TS and Diff-TS.
• Trained models will be saved in ./ModelSave/{dataset}/.

Testing

• Run validation.py for testing.
• The compressed results will be saved in ./SimpTraj.
• Query-related files will be saved in ./Val.

Acknowledgements

• https://github.com/zhengwang125/Query-TS
• https://github.com/Shark-NLP/DiffuSeq