This is a graph-based approach, which focusses on modeling the hierarchical spatial structure and improving the performance of long trajectory similarity computation.
Pytorch, Numpy, Yaml, Dgl, Networkx, Pickle, Scipy, Tensorboard, Tqdm, trajectory_distance
Please create 2 empty folders:
-
data: Path of the original data which is organized to a trajectory list. Each trajectory in it is a list of coordinate tuples (lon, lat). -
model/wts: It is used for placing the best TrajGAT model parameters of training.
Due to the file size limit, we put the dataset on other sites. Please first download the data and put it in data folder. The long trajectory dataset of Porto can be download at: https://drive.google.com/drive/folders/1hORrqGXXPZWiQXKVzAj0EFU6CYgIgeHd?usp=sharing
To train TrajGAT model, run the following command:
python main.py --config=model_config.yaml --gpu=0It trains TrajGAT under the supervision of metric distance. The parameters of TrajGAT can be modified in model_config.yaml.