Temporal Fusion Transformer for Interpretable Multi-horizon Forecasting
Attention-based architecture for multi-horizon time series forecasting with interpretable outputs.
Verdict
Strong performance on financial time series. Attention mechanism helps identify important features. Computationally expensive but worth it for complex forecasting tasks.
Practical Use
Excellent for multi-horizon forecasting problems with mixed input types (static covariates, known future inputs, and historical time series). The interpretable attention mechanism reveals which features drive predictions at each time horizon. Best suited for problems requiring both accuracy and explainability, though training can be computationally intensive for very long sequences or high-dimensional feature sets.
Paper
https://arxiv.org/abs/1912.09363Bryan Lim, Sercan O. Arik, Nicolas Loeff, Tomas Pfister. International Journal of Forecasting, 2021
License
Custom. Reference code (google-research/google-research, tft): Apache-2.0. Paper: arXiv non-exclusive distribution license (free to read; no reuse license).
Installation
pip install pytorch-forecasting