Commit 49107dda authored by Atreya Majumdar's avatar Atreya Majumdar
Browse files

Added detail to huber loss comment

parent 169805fd
Loading
Loading
Loading
Loading
+5 −1
Original line number Diff line number Diff line
@@ -42,7 +42,11 @@ class L1Loss(Loss):


class HuberLoss(Loss):
  """Modified version of L1 Loss, also known as Smooth L1 loss. Less sensitive to small errors, linear for larger errors."""
  """Modified version of L1 Loss, also known as Smooth L1 loss.
  Less sensitive to small errors, linear for larger errors.
  Huber loss is generally better for cases where are are both large outliers as well as small, as compared to the L1 loss.
  By default, Delta = 1.0 and reduction = 'none'.
  """

  def _compute_tf_loss(self, output, labels):
    import tensorflow as tf