STGFORMER_PRETRAIN_IMPUTE_NORM_METR-LA

Spatial-Temporal Graph Transformer (Pretrain Impute Norm) - METR-LA

Spatial-Temporal Graph Transformer (Pretrain Impute Norm) (STGFORMER_PRETRAIN_IMPUTE_NORM) trained on METR-LA dataset for traffic speed forecasting.

Model Description

STGFormer with masked node pretraining + imputation on NORMALIZED data (per-timestep -> per-node curriculum)

Dataset

METR-LA: Traffic speed data from highway sensors.

Usage

from utils.stgformer import load_from_hub

# Load model from Hub
model, scaler = load_from_hub("METR-LA", hf_repo_prefix="STGFORMER_PRETRAIN_IMPUTE_NORM")

# Get predictions
from utils.stgformer import get_predictions
predictions = get_predictions(model, scaler, test_dataset)

Training

Model was trained using the STGFORMER_PRETRAIN_IMPUTE_NORM implementation with default hyperparameters.

Citation

If you use this model, please cite the original STGFORMER_PRETRAIN_IMPUTE_NORM paper:

@inproceedings{lan2022stgformer,
  title={STGformer: Spatial-Temporal Graph Transformer for Traffic Forecasting},
  author={Lan, Shengnan and Ma, Yong and Huang, Weijia and Wang, Wanwei and Yang, Hui and Li, Peng},
  booktitle={IEEE Transactions on Neural Networks and Learning Systems},
  year={2022}
}

License

This model checkpoint is released under the same license as the training code.