Add comments for readability (pytorch#2005)

Author: yunjey

torch/nn/functional.py

@@ -663,6 +663,9 @@ def cross_entropy(input, target, weight=None, size_average=True, ignore_index=-1
  over observations for each minibatch. However, if the field
  sizeAverage is set to False, the losses are instead summed
  for each minibatch.
+ ignore_index (int, optional): Specifies a target value that is ignored
+ and does not contribute to the input gradient. When size_average is
+ True, the loss is averaged over non-ignored targets.
  """
  return nll_loss(log_softmax(input), target, weight, size_average, ignore_index)
 

torch/nn/modules/loss.py

@@ -409,10 +409,19 @@ class CrossEntropyLoss(_WeightedLoss):
 
  The losses are averaged across observations for each minibatch.
 
+ Args:
+ weight (Tensor, optional): a manual rescaling weight given to each class.
+ If given, has to be a Tensor of size "nclasses"
+ size_average (bool, optional): By default, the losses are averaged over observations for each minibatch.
+ However, if the field size_average is set to False, the losses are
+ instead summed for each minibatch.
+ ignore_index (int, optional): Specifies a target value that is ignored
+ and does not contribute to the input gradient. When size_average is
+ True, the loss is averaged over non-ignored targets.
+
  Shape:
  - Input: :math:`(N, C)` where `C = number of classes`
- - Target: :math:`(N)` where each value is `0 <= targets[i] <= C-1`
-
+ - Target: :math:`(N)` where each value is `0 <= targets[i] <= C-1
  """
 
  def __init__(self, weight=None, size_average=True, ignore_index=-100):