I wonder how different stages of results are ensembled in this method. This part of the code is not given, which should be very important according to the paper. Even in the evaluation and test stage, the loss is an average of loss in different stages, but not the loss of a final estimated result. I don't think this is reasonable.
I wonder how different stages of results are ensembled in this method. This part of the code is not given, which should be very important according to the paper. Even in the evaluation and test stage, the loss is an average of loss in different stages, but not the loss of a final estimated result. I don't think this is reasonable.