Cross-city Few-shot Traffic Forecasting via Traffic Pattern Bank

[CIKM 2023] In this repository, we presents the code of "Cross-city Few-shot Traffic Forecasting via Traffic Pattern Bank" (TPB).

Data

The data is in https://drive.google.com/drive/folders/1UrKTgR27YmP9PjJ-FWv4SCDH3zUxtc5R?usp=share_link. Please download it and save them in ./data

Environment

The code is implemented in pytorch 1.10.0, CUDA version 11.3, python 3.7.0.

pip install torch==1.10.0+cu113 -f https://download.pytorch.org/whl/torch_stable.html

Reproducibility

The default configs of the four datasets are set in ./config. To reproduce the result, please run following command:

bash train.sh

or run the experiment on specific dataset (PEMS-BAY as an example):

CUDA_VISIBLE_DEVICES=0 nohup python -u train.py  --config_file ./configs/config_pems.yaml > train_pems.out 2>&1 &

Pre-trained stuff

The pre-trained patch encoder and traffic pattern bank is contained in this repository. The pre-trained patch encoder is in ./save/pretrain_model and the traffic pattern bank is in ./pattern.

You can also pre-train and generate traffic pattern bank on your own by:

# Pre-train
python -u ./pretrain.py --test_dataset ${test_dataset} --data_list ${data_list}
wait
python -u ./patch_devide.py --test_dataset ${test_dataset} --data_list ${data_list}
wait
python -u ./pattern_clustering.py --data_list $data_list --test_dataset ${test_dataset} --sim ${sim} --K ${K}

${data_list} is the source data. For example, if you want to pre-train the encoder in Chengdu, METR-LA and PEMS-BAY, then ${data_list} is chengdu_metr_pems.

${test_dataset} is the dataset you want to build target data on. If you want to build target data on Shenzhen then the ${test_dataset} is shenzhen.

${sim} and ${K} are the clustering hyper-parameter. You can set them by your own.

Code Error to be corrected

There is something wrong in your code.
When I try to run the code, the following problem could not be eliminated:

Traceback (most recent call last):
  File "train.py", line 128, in <module>
    model_loss ,mse_loss, rmse_loss, mae_loss, mape_loss = rep_model.meta_train_revise(data_spt, matrix_spt, data_qry, matrix_qry)
  File "/root/autodl-nas/TPB/./model/Meta_Models/rep_model_final.py", line 251, in meta_train_revise
    out, y, meta_graph = self.model(data_spt[i], A_gnd)
  File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
    return forward_call(*input, **kwargs)
  File "/root/autodl-nas/TPB/./model/Meta_Models/rep_model_final.py", line 114, in forward
    raw_emb, Ax = STmodel(raw_x,A_list)
  File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
    return forward_call(*input, **kwargs)
  File "/root/autodl-nas/TPB/./model/Meta_Models/reconstruction.py", line 384, in forward
    gate = self.gate_convs[i](residual)
  File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl
    return forward_call(*input, **kwargs)
  File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 302, in forward
    return self._conv_forward(input, self.weight, self.bias)
  File "/root/miniconda3/lib/python3.8/site-packages/torch/nn/modules/conv.py", line 298, in _conv_forward
    return F.conv1d(input, weight, bias, self.stride,
RuntimeError: Expected 2D (unbatched) or 3D (batched) input to conv1d, but got input of size: [16, 32, 207, 13]

Your work is extremely valuable to me. I would greatly appreciate it if you could make an effort to resolve the bug.

zhyliu00 / tpb Goto Github PK

tpb's Introduction

Cross-city Few-shot Traffic Forecasting via Traffic Pattern Bank

Data

Environment

Reproducibility

Pre-trained stuff

tpb's People

Contributors

Stargazers

Watchers

Forkers

tpb's Issues

Recommend Projects

Recommend Topics

Recommend Org