jingweizuo / gcn-m

Repository for the paper "Graph Convolutional Networks for Traffic Forecasting with Missing Values" in DMKD'22

GCN-M: Graph Convolutional Networks for Traffic Forecasting with missing values

This is the companion repository for our paper titled Graph Convolutional Networks for Traffic Forecasting with missing values published in Data Mining and Knowledge Discovery and also available on ArXiv.

Requirements:

• matplotlib == 3.2.1
• numpy == 1.19.2
• pandas == 0.25.1
• scikit_learn == 0.21.2
• torch == 1.6.0
• tensorwatch == 0.9.1

Dependencies can be installed using the following command:

pip install -r requirements.txt

Data

Step1:

• Download METR-LA and PEMS-BAY data from Google Drive or Baidu Yun links provided by DCRNN.
• Put the downloaded data into the repository mentioned in "config/DATASET.conf"

Step2: Preprocess raw data

python data/generate_dated_data.py

Usage

python main.py --config CONFIG_FILE --itr NBR_ITERATION

Citation

If you find this repository useful in your research, please consider citing the following paper:

@article{zuo2023graph,
  title     = {Graph convolutional networks for traffic forecasting with missing values},
  author    = {Zuo, Jingwei and Zeitouni, Karine and Taher, Yehia and Garcia-Rodriguez, Sandra},
  journal   = {Data Mining and Knowledge Discovery},
  volume    = {37},
  number    = {2},
  pages     = {913--947},
  year      = {2023},
  publisher = {Springer}
}

gcn-m's Issues

error when run the GCN-M model

In the definition of GCNMdynamic model, define the gated_conv as:
self.gate_convs.append(nn.Conv1d(in_channels=residual_channels, out_channels=dilation_channels, kernel_size=(1, kernel_size), dilation=new_dilation))

when run the model in metr_la dataset, get an error when calculate gate = self.gate_convs[i](residual) # kernel=(1,2)
RuntimeError: Expected 2D (unbatched) or 3D (batched) input to conv1d, but got input of size: [32, 32, 207, 13]

I think the residual got the correct shape of (B, residual_size, D, F) as noted. Is this a problem caused by the improper vision of torch?
Could you please help me solve this problem, I will appreciate any assistance you can provide.

No module named encoder, decoder, attn and embed

Thank you for your great contributions! I failed to run the code because no modules named encoder, decoder, attn, and embed. Could you please provide these files? Thank you for your patience!