safarzadeh-reza / T-GCN-PyTorch

A PyTorch implementation of T-GCN

T-GCN-PyTorch

This is a PyTorch implementation of T-GCN in the following paper: T-GCN: A Temporal Graph Convolutional Network for Traffic Prediction.

A stable version of this repository can be found at the official repository.

Notice that the original implementation is in TensorFlow, which performs a tiny bit better than this implementation for now.

Requirements

• numpy
• matplotlib
• pandas
• torch
• pytorch-lightning>=1.3.0
• torchmetrics>=0.3.0
• python-dotenv

Model Training

# GCN
python main.py --model_name GCN --max_epochs 3000 --learning_rate 0.001 --weight_decay 0 --batch_size 64 --hidden_dim 100 --settings supervised --gpus 1
# GRU
python main.py --model_name GRU --max_epochs 3000 --learning_rate 0.001 --weight_decay 1.5e-3 --batch_size 64 --hidden_dim 100 --settings supervised --gpus 1
# T-GCN
python main.py --model_name TGCN --max_epochs 3000 --learning_rate 0.001 --weight_decay 0 --batch_size 32 --hidden_dim 64 --loss mse_with_regularizer --settings supervised --gpus 1

You can also adjust the --data, --seq_len and --pre_len parameters.

Run tensorboard --logdir lightning_logs/version_0 to monitor the training progress and view the prediction results.