weights acts as a coefficient for the loss. If a scalar is provided,
then the loss is simply scaled by the given value. If weights is a
tensor of size [batch_size], then the loss weights apply to each
If label_smoothing is nonzero, smooth the labels towards 1/num_classes:
new_onehot_labels = onehot_labels * (1 - label_smoothing)
+ label_smoothing / num_classes
[batch_size, num_classes] logits outputs of the network .
[batch_size, num_classes] one-hot-encoded labels.
Coefficients for the loss. The tensor must be a scalar or a tensor
of shape [batch_size].
If greater than 0 then smooth the labels.
the scope for the operations performed in computing the loss.
A scalar Tensor representing the mean loss value.
If the shape of logits doesn't match that of onehot_labels
or if the shape of weights is invalid or if weights is None.