TiDE

A unofficial pytorch implementation of "Long-term Forecasting with TiDE: Time-series Dense Encoder" and its sample code of applications

Link to paper: Long-term Forecasting with TiDE: Time-series Dense Encoder

Official Code written by Tensorflow: Code

Usage

1. Config model

   edit the Tide_{dataset_name}.yaml file in the dir of config/TiDE/

2. Train

   python main_TiDE.py --dataset 'dataset_name'

Details

1. The dimension of Attribute is [num_nodes, num_hidden_attribute]. Here num_nodes is equal to the number of nodes in the original dataset, i.e., 7 for the ETT dataset, 862 for the traffic dataset. num_hidden_attribute=16.
2. Dynamic Covariates' dimension is equal to 7 or 25, depending on whether you use holiday features or not, we use freq: 'B' to represent dynamic covaraites without holiday features and freq: 'S' to represent dynamic covaraites with holiday features