---
tags:
- traffic-forecasting
- time-series
- graph-neural-network
- mtgnn
datasets:
- metr-la
---

# MTGNN Model - METR-LA

Multivariate Time Series Forecasting with Graph Neural Networks (MTGNN) trained on METR-LA dataset for traffic speed forecasting.

## Model Description

This model uses a graph neural network architecture that combines:
- Graph learning to automatically discover spatial dependencies
- Temporal convolution for modeling temporal patterns
- Mix-hop propagation for capturing multi-scale spatial patterns

## Evaluation Metrics

- **Validation MAE**: 2.7572
- **Test MAE (15 min)**: 6.6690
- **Test MAPE (15 min)**: 0.2444
- **Test RMSE (15 min)**: 12.8614


## Dataset

**METR-LA**: Traffic speed data from highway sensors.

## Usage

```python
from utils.mtgnn import load_from_hub

# Load model from Hub
model = load_from_hub("METR-LA")

# Get predictions
import numpy as np
x = np.random.randn(10, 2, 207, 12)  # (batch, features, nodes, seq_len)
predictions = model.predict(x)
```

## Training

Model was trained using the MTGNN implementation with default hyperparameters.

## Citation

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

```bibtex
@inproceedings{wu2020connecting,
  title={Connecting the Dots: Multivariate Time Series Forecasting with Graph Neural Networks},
  author={Wu, Zonghan and Pan, Shirui and Long, Guodong and Jiang, Jing and Chang, Xiaojun and Zhang, Chengqi},
  booktitle={Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining},
  pages={753--763},
  year={2020}
}
```

## License

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