Hello all,
I am trying to implement a custom loss function, but with a condition
def ReverseSmoothL1Loss(output, target):
absolute_errors = torch.abs(output - target)
absolute_errors_squared = torch.squared(absolute_errors)
loss = torch.where(absolute_errors > 1, absolute_errors_squared, absolute_errors)
return torch.mean(loss)
I am getting valid results, but is this implementation valid for autograd?