Open rginjapan opened 3 months ago
However, in the paper you said "We use the historical ground-truth data from the past 12 steps to forecast one future time step", the pred is the same size as input in your implementation, so how and where to get the one future time prediction?
How can I change this model to multi-ouputs(1 to k moment) to one output (k+1 moment)? using the average of dim=1 to convert the prediction of (b,12,184) to (b,1,184)??
Is it possible to use several previous inputs to predict only next one moment? such as seq_len = 12, but pred_len =1, currently the code is only work with seq_len == pred_len