About Code release for "Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting" (NeurIPS 2021), https://arxiv.org/abs/2106.13008
Yes, in Auto-Correlation we don't need mask.
But if you want to process time series with missing data, you may need this parameter to mark the meaningless tokens.
The decoder has no implementation of mask mechanism. So are those parameters like 'x_mask' are redundant?