MengzhangLI/STFGNN

temporal denpendencies

wanzhixiao opened this issue · 1 comments

Hi,thanks for your excellent work!
I have a question, is the performance of Gated CNN berter than TCN?

Hi,thanks for your excellent work!
I have a question, is the performance of Gated CNN berter than TCN?

I think TCN can be added in this module only with some big modifications on STFGNN. The reasons I use Gated CNN are:
(1) 1D CNN with dilation rate=3 could be adaptively added into STFGN Layer.
(2) Gated Mechanism could really help model performance.

TCN is a great work, GraphWaveNet proves its performance. However, it can hardly beat RNN/LSTM/transformer in many cases.