RATD

This repository contains the experiments in the Neurips 2024 paper "Retrieval-Augmented Diffusion Models for Time Series Forecasting" by Jingwei Liu, Ling Yang, Hongyan Li and Shenda Hong.

Requirement

Please install the packages in requirements.txt

Notice

This version is not the final version of our code, we will update the full version ASAP.

Data

The electricity dataset can be found at ./data/ts2vec

Experiments

Retrieval

We use the TCN as our encoder, the code can be found at ./TCN-master.

python retrieval.py --type encode

To save the references, you can run

python retrieval.py --type retrieval

Training and forecasting for the electricity dataset

python exe_forecasting.py --datatype electricity

Acknowledgements

Codes are based on CSDI