This repository contains the code used in paper

Requirements

• Ubuntu OS
• Python 3.9.13 (tested)
• pip install -r requirements.txt
• Download arbitrary dataset (e.g., Porto) to "./data/{dataset_name}/{dataset_name}.pkl", convert it to a DataFrame containing columns ["Traj_ID", "Timestamps", "Locations", "Length"].
• Download the osm file (e.g., "portugal-latest.osm.pbf") to "./data/{dataset_name}/meta/"

Map-matching

1. Get the .osm file of the city

   python map-matching/osm_convert.py

2. Get the road network files for the city (such as "porto_nodes.csv")

   python map-matching/osm2roadnetwork.py

3. Get the map-matched results for trajectories in "porto.pkl".

   python map-matching/HMMM.py

   Note that even with multiprocessing, it will take a long time.

Preprocessing

1. Data augmentation for pre-training

   python preprocess/augmentation.py

2. Extract road segments features

   python preprocess/rs_extract.py

3. Extract trajectory features.

   python preprocess/feature_extract.py

Pre-training

1. Pre-training on a large dataset, which might take a long time.

   python pretrain.py

Fine-tuning

1. Generate the fine-tuning dataset

   python dataset_preparation.py

2. Fine-tune the pre-trained encoder

   python finetune_SL.py

   If you want to train from scratch, set "self.load_pretrain" of Config() in config.py to False.