This repository is the official implementation of Towards Robust Real-World Multivariate Time Series Forecasting: A Unified Framework for Dependency, Asynchrony, and Missingness.
To install requirements:
pip install -r requirements.txt
To train the model in the paper, run this command:
bash ./scripts_practical/ChannelTokenFormer/CTF_ETT1_practical.sh
These scripts include evaluation also.
To evaluate model on SolarWind-missing scenarios, you should first train and save a checkpoint with the practical setting, then run missing-scenario inference.
- Train with SolarWind practical setting and generate checkpoint files:
bash ./scripts_practical/ChannelTokenFormer/CTF_SW_practical.sh
-
Confirm the saved checkpoint directory from the training output (or your configured checkpoints path).
-
Set that checkpoint directory in the missing evaluation script/config, then run:
bash ./scripts_practical/ChannelTokenFormer/CTF_SW_missing.sh
CTF_SW_missing.sh requires a valid pretrained checkpoint path from Step 1.
We appreciate the following GitHub repos a lot for their valuable code and efforts.
- Time-Series-Library (https://github.com/thuml/Time-Series-Library)
- CrossGNN (https://github.com/hqh0728/CrossGNN)