Any easy workaround for using weighted smooth_l1 loss for different output units?

Current it seems only support sample-wise weighting.
How about having an element wise option for loss function?

there isn’t an easy work-around. you can call the loss in a for-loop over the mini-batch.

1 Like