Merge pull request #48 from MaxTuecke/keras_documentation

Added documentation to custom_keras_objects

docs/servable-types.rst

@@ -112,7 +112,30 @@ can be created from this model file and names for the output classes:
 
 .. code-block:: python
 
-    model info = KerasModel.create_model('model.h5', ["y"])
+    model_info = KerasModel.create_model('model.h5', ["y"])
+
+Models with weights and architecture as separate files can be described using:
+
+.. code-block:: python
+
+	model_info = KerasModel.create_model('model.h5', ["y"], arch_path='arch.json')
+
+Keras also allows users to add their own custom layers to their models for any custom operation
+that has trainable weights. Use this when the Keras Lambda layer does not apply. In Keras, 
+these layers can be added when loading the model:
+
+.. code-block:: python
+
+	model = load_model('model.h5', custom_objects={'CustomLayer': CustomLayer})
+
+Adding custom layers to a DLHub description can be achived with the ``add_custom_object`` method, which takes the name
+and class of the custom layer:
+
+.. code-block:: python
+
+	model_info.add_custom_object('CustomLayer', CustomLayer)
+
+See more info on creating custom Keras layers `here <https://keras.io/layers/writing-your-own-keras-layers/>`_.
 
 The DLHub SDK reads the architecture in the HDF5 file and determines the inputs
 and outputs automatically: