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
1.37k stars 248 forks source link

Question: is patchtst_supervised.py in self-supervised model same as supervised model? #62

Closed koseoyoung closed 1 year ago

koseoyoung commented 1 year ago

Hi, I really appreciate your help so far!

I have one more question about the codebase: What is the difference between patchtst_supervised.py in the self-supervised folder (https://github.com/yuqinie98/PatchTST/blob/main/PatchTST_self_supervised/patchtst_supervised.py#L104) and PatchTST training in the supervised folder (https://github.com/yuqinie98/PatchTST/blob/main/PatchTST_supervised/run_longExp.py#L140C1-L140C1) in terms of the way of training? Can I understand both training scripts are doing the same thing?

Thanks!

yuqinie98 commented 1 year ago

Hi, generally speaking they are very similar, some main differences: 1. the supervised code is modified from previous repo like Autoformer, FEDformer, while the self-supervised code is adapted from tsai repo. 2. The supervised task is doing prediction while the self-supervised one is reconstructing the masked patches.

koseoyoung commented 1 year ago

I see, thanks for sharing. :)