def sparse_softmax_cross_entropy(
labels,
logits,
weights=1.,
scope=None,
loss_collection=ops.GraphKeys.LOSSES,
reduction=Reduction.SUM_BY_NONZERO_WEIGHTS,
):
if labels is None: raise ValueError("labels must not be None.")
if logits is None: raise ValueError("logits must not be None.")
normalization = None
if reduction == Reduction.NONE: normalization = 'UNIT'
elif reduction == Reduction.MEAN: normalization = 'FULL'
elif reduction == Reduction.SUM_BY_NONZERO_WEIGHTS or \
reduction == Reduction.SUM_OVER_NONZERO_WEIGHTS:
normalization = 'NONE'
elif reduction == Reduction.SUM_OVER_BATCH_SIZE:
normalization = 'BATCH_SIZE'
loss = _ops.SparseSoftmaxCrossEntropy(
[logits, labels],
normalization=normalization,
name=scope,
)
if weights != 1.0: loss = weights * loss
ops.add_to_collection(loss_collection, loss)
return loss
def sparse_softmax_cross_entropy_with_logits(logits,
labels,
dim=-1,
name=None):
if dim == -1: dim = 1
return ops.SparseSoftmaxCrossEntropy([logits, labels],
axis=dim,
normalization='UNIT',
name=name)
def sparse_softmax_cross_entropy_with_logits(
_sentinel=None,
labels=None,
logits=None,
dim=-1,
name=None,
):
return _ops.SparseSoftmaxCrossEntropy(
[logits, labels],
axis=dim,
normalization='UNIT',
name=name,
)
def Setup(self, bottom):
super(SoftmaxWithLossLayer, self).Setup(bottom)
loss = ops.SparseSoftmaxCrossEntropy(bottom, **self._param)
if self._loss_weight is not None: loss *= self._loss_weight
return loss
def LayerSetup(self, bottom):
loss = _ops.SparseSoftmaxCrossEntropy(bottom, **self.arguments)
if self._loss_weight is not None: loss *= self._loss_weight
return loss