add doc and API.spec. test=develop

paddle/fluid/API.spec

@@ -220,6 +220,7 @@ paddle.fluid.layers.py_func (ArgSpec(args=['func', 'x', 'out', 'backward_func',
 paddle.fluid.layers.psroi_pool (ArgSpec(args=['input', 'rois', 'output_channels', 'spatial_scale', 'pooled_height', 'pooled_width', 'name'], varargs=None, keywords=None, defaults=(None,)), ('document', '1546136806fef5c08f6918544bd9151d'))
 paddle.fluid.layers.teacher_student_sigmoid_loss (ArgSpec(args=['input', 'label', 'soft_max_up_bound', 'soft_max_lower_bound'], varargs=None, keywords=None, defaults=(15.0, -15.0)), ('document', '2f6ff96864054a31aa4bb659c6722c99'))
 paddle.fluid.layers.huber_loss (ArgSpec(args=['input', 'label', 'delta'], varargs=None, keywords=None, defaults=None), ('document', '431a4301c35032166ec029f7432c80a7'))
+paddle.fluid.layers.kldiv_loss (ArgSpec(args=['x', 'target', 'reduction', 'name'], varargs=None, keywords=None, defaults=('mean', None)), ('document', '26e3842d408b0af4653433ce1591a473449a78f6'))
 paddle.fluid.layers.tree_conv (ArgSpec(args=['nodes_vector', 'edge_set', 'output_size', 'num_filters', 'max_depth', 'act', 'param_attr', 'bias_attr', 'name'], varargs=None, keywords=None, defaults=(1, 2, 'tanh', None, None, None)), ('document', '34ea12ac9f10a65dccbc50100d12e607'))
 paddle.fluid.layers.data (ArgSpec(args=['name', 'shape', 'append_batch_size', 'dtype', 'lod_level', 'type', 'stop_gradient'], varargs=None, keywords=None, defaults=(True, 'float32', 0, VarType.LOD_TENSOR, True)), ('document', '33bbd42027d872b3818b3d64ec52e139'))
 paddle.fluid.layers.open_files (ArgSpec(args=['filenames', 'shapes', 'lod_levels', 'dtypes', 'thread_num', 'buffer_size', 'pass_num', 'is_test'], varargs=None, keywords=None, defaults=(None, None, 1, None)), ('document', 'b1ae2e1cc0750e58726374061ea90ecc'))

paddle/fluid/operators/kldiv_loss_op.cc

@@ -89,6 +89,24 @@ class KLDivLossOpMaker : public framework::OpProtoAndCheckerMaker {
          This operator calculates the Kullback-Leibler divergence loss
          between Input(X) and Input(Target).
 
+         KL divergence loss calculates as follows:
+
+         $$l(x, y) = y * (\log y - x)$$
+
+         While :attr:`reduction` is :attr:`none`, output loss is in
+         same shape with Input(X), loss in each point is calculated 
+         seperately and no reduction applied.
+         
+         While :attr:`reduction` is :attr:`mean`, output loss in in
+         shape of [1] and loss value is the mean value of all losses.
+         
+         While :attr:`reduction` is :attr:`sum`, output loss in in
+         shape of [1] and loss value is the sum value of all losses.
+         
+         While :attr:`reduction` is :attr:`batchmean`, output loss in 
+         in shape of [1] and loss value is the sum value of all losses
+         divided by batch size.
+         
          )DOC");
   }
 };

python/paddle/fluid/layers/nn.py

@@ -186,6 +186,7 @@ __all__ = [
     'psroi_pool',
     'teacher_student_sigmoid_loss',
     'huber_loss',
+    'kldiv_loss',
     'tree_conv',
 ]
 
@@ -10588,6 +10589,38 @@ def huber_loss(input, label, delta):
     return out
 
 
+@templatedoc()
+def kldiv_loss(x, target, reduction='mean', name=None):
+    """
+    ${comment}
+
+    Args:
+        x (Variable): ${x_comment}
+        target (Variable): ${target_comment}
+        reduction (Variable): ${reduction_comment}
+        name (str, default None): The name of this layer.
+
+    Returns:
+        kldiv\_loss (Variable): The KL divergence loss.
+
+    Examples:
+        .. code-block:: python
+
+            x = fluid.layers.data(name='x', shape=[4,2,2], dtype='float32')
+            target = fluid.layers.data(name='target', shape=[4,2,2], dtype='float32')
+            loss = fluid.layers.kldiv_loss(x=x, target=target, reduction='batchmean')
+    """
+    helper = LayerHelper('kldiv_loss', **locals())
+    loss = helper.create_variable_for_type_inference(dtype=x.dtype)
+    helper.append_op(
+        type='kldiv_loss',
+        inputs={'X': x,
+                'Target': target},
+        outputs={'Loss': loss},
+        attrs={'reduction': reduction})
+    return loss
+
+
 @templatedoc()
 def tree_conv(nodes_vector,
               edge_set,

python/paddle/fluid/tests/unittests/test_layers.py

@@ -1046,6 +1046,15 @@ class TestBook(unittest.TestCase):
             out = layers.spectral_norm(weight, dim=1, power_iters=1)
             self.assertIsNotNone(out)
 
+    def test_kldiv_loss(self):
+        program = Program()
+        with program_guard(program):
+            x = layers.data(name='x', shape=[32, 128, 128], dtype="float32")
+            target = layers.data(
+                name='target', shape=[32, 128, 128], dtype="float32")
+            loss = layers.kldiv_loss(x=x, target=target, reduction='batchmean')
+            self.assertIsNotNone(loss)
+
         print(str(program))
 
     def test_shuffle_channel(self):