ERF: Worked on examples.

sphinx/source/examples_src/bimdpinv_src/bimdpinv.rst

@@ -0,0 +1,39 @@
+.. _examples_bimdpinv:
+
+BiFlow Inverse
+==============
+::
+
+    """
+    Example on how calculate the inverse of a BiFlow.
+
+    A BiFlow does inherit the 'inverse' method from the mdp.Flow class, but
+    this doesn't support BiMDP features like messages. If these features are
+    needed then one can use an alternative way to calucalte the inverse, which
+    is presented here.
+    """
+
+    import numpy as np
+    import bimdp
+
+    # create a simple pointless flow
+    pca_node = bimdp.nodes.PCABiNode()
+    sfa_node = bimdp.nodes.SFABiNode()
+    flow = pca_node + sfa_node
+    x = np.random.random((50,5))
+    flow.train(x)
+
+    x = np.random.random((3,5))
+    y = flow.execute(x)
+
+    # the target value 1 is the absolute index of the sfa_node,
+    # alternatively one could have used a node_id
+    inv_x, _ = flow.execute(y, {"method": "inverse"}, 1)
+    #_, (inv_x, _) = bimdp.show_execution(flow, y, {"method": "inverse"}, 1)
+    assert(np.all(np.abs(x - inv_x) < 0.0000001))
+
+    # compare the result to the standard inverse
+    inv2_x = flow.inverse(y)
+    assert(np.all(np.abs(inv2_x - inv_x) < 0.0000001))
+
+    print "done."

sphinx/source/examples_src/examples.rst

@@ -8,6 +8,7 @@ Examples
    logmap_src/logmap.rst
    lle_src/lle.rst
    gng_src/gng.rst
+   bimdpinv_src/bimdpinv.rst
    mlpbackprop_src/mlpbackprop.rst
    gradient_src/gradient.rst
 
@@ -23,6 +24,8 @@ applications:
 
 The following examples use and illustrate BiMDP:
 
+* :ref:`examples_bimdpinv` - A simple example on the alternative mechanism
+  to inverse a BiFlow.
 * :ref:`examples_mlpbackprop` - Implement backpropagation for a multi layer
   perceptron.
 * :ref:`examples_graddescent` - Use Newton's method for gradient descent

sphinx/source/examples_src/mlpbackprop_src - Kopie (2)/mlpbackprop.rst

@@ -0,0 +1,8 @@
+.. _examples_mlpbackprop:
+
+Backpropagation
+===============
+
+This example shows how to implement the backpropagation learning method in the
+BiMDP framework. It uses this to implement a multi layer perceptron.
+The example code is available in the MDP examples repository.