search

yuqinie98 / PatchTST

An offical implementation of PatchTST: "A Time Series is Worth 64 Words: Long-term Forecasting with Transformers." (ICLR 2023) https://arxiv.org/abs/2211.14730
Apache License 2.0
1.52k stars 262 forks source link

How to pachify non-uniform data? #4

Closed panjiashu closed 1 year ago

panjiashu commented 1 year ago

Thank you for the great work. I have a small question. Many time series in the real world are non-uniform, but patching seems to require uniform sampling. How to deal with it?

namctin commented 1 year ago

Hi @panjiashu, thanks for your great question! Yes, non-uniformity often occurs with with time series and we have not found an efficient solution to handle this scenario with patching yet.

yuqinie98 commented 1 year ago

Hi @panjiashu, thanks for your question! As @namctin suggested, we haven't take this into consideration yet. But if you want to use PatchTST on non-uniform time series, one way is to re-sample it to make it uniform. Based on table 1, this hopefully won't hurt the performance too much.