Skip to content

Latest commit

 

History

History
20 lines (13 loc) · 685 Bytes

README.md

File metadata and controls

20 lines (13 loc) · 685 Bytes

tft-forecast

Time Series Forecasting with Temporal Fusion Transformers

Based on: Lim, Bryan & Arik, Sercan & Loeff, Nicolas & Pfister, Tomas. (2019). Temporal Fusion Transformers for Interpretable Multi-horizon Time Series Forecasting
Paper: https://arxiv.org/pdf/1912.09363.pdf

Proposal: https://github.com/damu4/tft-forecast/raw/main/proposal.pdf

Project goals:

  1. Reproduce the experiment with volatility dataset.

  2. In the original paper data used from 2000-01-03 to 2019-06-28. I will train model on data to 2021-05-01

  3. Tune model's hyperparameters.

  4. Experiments with changing the model:
    a. Change Softmax to SM-Taylor softmax
    b. Add linear layer to GRU