Added dump/load into library documentation

doc/library/index.txt

@@ -20,6 +20,7 @@ Types and Ops that you can use to build and compile expression graphs.
    sparse/sandbox
    scalar/index
    gof/index
+   misc/pkl_utils
    scan
    sandbox/index
    typed_list

doc/library/misc/pkl_utils.txt

+
+.. _libdoc_misc:
+
+================================================
+:mod:`misc.pkl_utils` - Tools for serialization.
+================================================
+
+.. autofunction:: theano.misc.pkl_utils.dump
+
+.. autofunction:: theano.misc.pkl_utils.load
+
+.. seealso::
+
+    :ref:`tutorial_loadsave`
+
+

doc/tutorial/loading_and_saving.txt

@@ -118,9 +118,18 @@ Robust Serialization
 ====================
 
 This type of serialization internally uses pickling but extracts values of all
-shared variables and saves it as numpy arrays. So it may be very hard to
-deserialize objects on a different version of Theano. It is useful when you
-would like to resume an experiment on a different kind of hardware.
+arrays (`CudaNdarray`, `ndarray`) and saves it as numpy arrays. So it may be
+very hard to deserialize objects on a different version of Theano.
+But tt is useful when you would like to resume an experiment on a different
+kind of hardware.
+
+Another use case of this type of serialization is to use parameters for another
+model after pretraining, fixing bugs or minor changes in it. You can still
+load the parameters manually with `numpy`:
+
+.. code-block:: python
+
+    numpy.load('model.zip')
 
 .. autofunction:: theano.misc.pkl_utils.dump
 

theano/misc/tests/test_pkl_utils.py

@@ -36,4 +36,3 @@ def test_dump_load_mrg():
         rng = load(f)
 
     assert type(rng) == MRG_RandomStreams
-