AdapGL

This is a Pytorch implemention of AdapGL.

Requirements

The model is implemented using python3 with dependencies specified in requirements.txt.

Traffic datasets

PeMSD4 and PeMSD8 datasets can be downloaded from PeMS-BAY with password "qhoa". Move them into data folder.

Model Training (for PeMSD4)

• AdapGL+ASTGCN

python ./main.py \
    --model_config_path ./config/train_pems04.yaml \
    --train_config_path ./config/train_config.yaml \
    --model_name AdapGLA \
    --num_epoch 5 \
    --num_iter 20 \
    --model_save_path ./model_states/AdapGLA_pems04.pkl \
    --max_graph_num 3

• AdapGL+DCRNN

python ./main.py \
    --model_config_path ./config/train_pems04.yaml \
    --train_config_path ./config/train_config.yaml \
    --model_name AdapGLD \
    --num_epoch 10 \
    --num_iter 15 \
    --model_save_path ./model_states/AdapGLD_pems04.pkl \
    --max_graph_num 3

• AdapGL+TGCN

python main.py \
    --model_config_path ./config/train_pems04.yaml \
    --train_config_path ./config/train_config.yaml \
    --model_name AdapGLT \
    --num_epoch 5 \
    --num_iter 20 \
    --model_save_path ./model_states/AdapGLT_pems04.pkl \
    --max_graph_num 3