The code for FilterTS: Comprehensive Frequency Filtering for Multivariate Time Series Forecasting
Please install the corresponding version of PyTorch via the official PyTorch website, then follow these steps to set up your environment.
pip install -r requirements.txtDownload the dataset from the following link and place it in the ./dataset/ directory.
(Please note, this dataset download link is provided by the author of iTransformer and is not associated with any author's address)
Run the shell script in the ./scripts folder, using the following command, for example:
sh ./scripts/FilterTS_electricity.sh
FilterTS demonstrates superior accuracy across eight multivariate long-term time series forecasting datasets compared to current state-of-the-art models in most cases.
During a recent code review we discovered that an unintended drop_last=True option was left active in the dataloader. Below are the results obtained after rerunning all experiments with drop_last=False. Across the majority of settings the differences are negligible, confirming that the original conclusions remain valid.
