Merge pull request #11560 from JiayiFeng/doc_non_layer_api

Doc of non layer api

python/paddle/fluid/average.py

@@ -36,6 +36,25 @@ def _is_number_or_matrix_(var):
 
 
 class WeightedAverage(object):
+    """
+    Calculate weighted average.
+
+    The average calculating is accomplished via Python totally. 
+    They do not change Paddle's Program, nor do anything to
+    modify NN model's configuration. They are completely 
+    wrappers of Python functions.
+
+    Examples:
+        .. code-block:: python
+            avg = fluid.average.WeightedAverage()
+            avg.add(value=2.0, weight=1)
+            avg.add(value=4.0, weight=2)
+            avg.eval()
+
+            # The result is 3.333333333.
+            # For (2.0 * 1 + 4.0 * 2) / (1 + 2) = 3.333333333
+    """
+
     def __init__(self):
         warnings.warn(
             "The %s is deprecated, please use fluid.metrics.Accuracy instead." %

python/paddle/fluid/backward.py

@@ -435,18 +435,65 @@ def _get_stop_gradients_(program):
 def append_backward(loss, parameter_list=None, no_grad_set=None,
                     callbacks=None):
     """
-    Append backward part to main_program
+    Append backward part to main_program.
 
-    Args:
-        loss(Variable): The variable generated by cost function.
-        parameter_list(list[string]): Parameters that need to be updated by
-            optimizer. If None, it means all parameters need to be updated.
-        no_grad_set(set): Variables that have no gradients in Block 0.
-            All variables with `step_gradient=True` from all blocks will be
-            automatically added.
+    A complete neural network training is made up of forward and backward 
+    propagation. However, when we configure a network, we only need to 
+    specify its forwrd part. The backward part is generated automatically 
+    according to the forward part by this function.
 
-    Return:
-        (list[(Variable,Variable)]): list of (parameter, gradient) pair.
+    In most cases, users do not need to invoke this function manually. It 
+    will be automatically invoked by the optimizer's `minimize` function.
+
+    Args:
+        loss(Variable): The loss variable of the network.
+        parameter_list(list[string]|None): Names of parameters that need 
+                                           to be updated by optimizers. 
+                                           If it is None, all parameters 
+                                           will be updated.
+                                           Default: None
+        no_grad_set(set|None): Variables in the Block 0 whose gradients 
+                               should be ignored. All variables with 
+                               `step_gradient=True` from all blocks will 
+                               be automatically added into this set.
+                               Default: None
+        callbacks(list[callable object]|None): The callbacks are used for 
+                                               doing some custom jobs during 
+                                               backward part building. All 
+                                               callable objects in it will 
+                                               be invoked once each time a 
+                                               new gradient operator is added 
+                                               into the program. The callable 
+                                               object must has two input 
+                                               parameters: 'block' and 'context'. 
+                                               The 'block' is the block which 
+                                               the new gradient operator will 
+                                               be added to. The 'context' is a 
+                                               map, whose keys are gradient 
+                                               variable names and values are 
+                                               corresponding original variables.
+                                               In addition to this, the 'context' 
+                                               has another special key-value pair: 
+                                               the key is string '__current_op_desc__' 
+                                               and the value is the op_desc of the 
+                                               gradient operator who has just 
+                                               triggered the callable object. 
+
+    Returns:
+        list[(Variable,Variable)]: Pairs of parameter and its 
+        corresponding gradients. The key is the parameter and the 
+        value is gradient variable.
+
+    Raises:
+        AssertionError: If `loss` is not an instance of Variable.
+
+    Examples:
+        .. code-block:: python
+
+            # network configuration code
+            # ...
+            avg_loss = fluid.layers.mean(loss)
+            param_grad_list = fluid.backward.append_backward(loss=avg_loss)
     """
     assert isinstance(loss, framework.Variable)