Align subclassing guide with docstring of Layer.add_loss() (#1144)

* Align subclassing guide with docstring of Layer.add_loss()

which uses `reduce_mean()`, not `reduce_sum()`, to achieve independence of batch sizes, consistent with Keras' default loss reduction by SUM_OVER_BATCH_SIZE. Along the way, clarify the difference between `add_loss` and `Loss`.

* Update making_new_layers_and_models_via_subclassing.py

Addresses review comment.

Author: arnoegw

guides/making_new_layers_and_models_via_subclassing.py

@@ -236,11 +236,14 @@ def __init__(self, rate=1e-2):
         self.rate = rate
 
     def call(self, inputs):
-        self.add_loss(self.rate * tf.reduce_sum(inputs))
+        self.add_loss(self.rate * tf.reduce_mean(inputs))
         return inputs
 
 
 """
+Notice that `add_loss()` can take the result of plain TensorFlow operations.
+There is no need to call a `Loss` object here.
+
 These losses (including those created by any inner layer) can be retrieved via
 `layer.losses`. This property is reset at the start of every `__call__()` to
 the top-level layer, so that `layer.losses` always contains the loss values