Forecasting head

[bwrooney82]

Hi ! Thank you for your interesting work. In some scenarios such as anomaly detection, how to modify the forecasting decoder into a reconstructed decoder while matching the dimensions of the input (B,M,L,N)?

[willzhang3]

The final latent representation (equation 12) by using t-PatchGNN as an encoder is (B, N, D), where N is the number of variables and D is the hidden dimension of the representation. This latent representation can be used by the task-specific decoder for other time series tasks, such as anomaly detection.