pytorch-forecasting icon indicating copy to clipboard operation
pytorch-forecasting copied to clipboard

Published 20 hours ago • verified publisher icon jdb78

Known vs. Unknown Variables

Open thiagopx opened this issue 2 years ago • 0 comments

Guys, I have a more conceptual issue on the TemporalFusionTransformer.

I am aware that the encoder and decoder must have the same input dimensionality. When I use TimeSeriesDataSet to create a dataset with 1 known cont. variable and 2 unknown cont. variables, it results in dimensionality 3. However, the unknown variables (2 dimensions) in the decoder should be disregarded since they are unknown in the future. How the implementation deals with this matter? I was wondering whether the decoder mask would do that, but I was not able to point that in the source code.

I appreciate any help with this issue.

thiagopx avatar Sep 01 '22 12:09 thiagopx