Fix to make example_crbm.py work with the modified code.

The examples belong to the version of Morb without fast PCD.
This is now also mentioned in the updated README.md.

README.md

@@ -4,6 +4,9 @@ This is a modified version of Morb. This version provides
  * persistent contrastive divergence (PCD) with fast weights;
  * a number of small changes and fixes.
 
+Note that the examples were written for the original version of Morb
+and may not directly work with this modified version.
+
 This is part of the code used for the convolutional classification RBMs in
 
 > Learning Features for Tissue Classification with the Classification

examples/example_crbm.py

@@ -37,21 +37,22 @@
 print ">> Constructing RBM..."
 rbm = rbms.BinaryBinaryCRBM(n_visible, n_hidden, n_context)
 initial_vmap = { rbm.v: T.matrix('v'), rbm.x: T.matrix('x') }
+pmap = rbm.pmap
 
 # try to calculate weight updates using CD-1 stats
 print ">> Constructing contrastive divergence updaters..."
-s = stats.cd_stats(rbm, initial_vmap, visible_units=[rbm.v], hidden_units=[rbm.h], context_units=[rbm.x], k=1)
+s = stats.cd_stats(rbm, initial_vmap, pmap, visible_units=[rbm.v], hidden_units=[rbm.h], context_units=[rbm.x], k=1)
 
 umap = {}
 for var in rbm.variables:
-    pu = var + 0.0005 * updaters.CDUpdater(rbm, var, s)
-    umap[var] = pu
+    pu = pmap[var] + 0.0005 * updaters.CDUpdater(rbm, var, s)
+    umap[pmap[var]] = pu
 
 print ">> Compiling functions..."
 t = trainers.MinibatchTrainer(rbm, umap)
 m = monitors.reconstruction_mse(s, rbm.v)
 mce = monitors.reconstruction_crossentropy(s, rbm.v)
-free_energy = T.mean(rbm.free_energy([rbm.h], s['data'])) # take the mean over the minibatch.
+free_energy = T.mean(rbm.free_energy([rbm.h], s['data'], pmap)) # take the mean over the minibatch.
 
 # train = t.compile_function(initial_vmap, mb_size=32, monitors=[m], name='train', mode=mode)
 train = t.compile_function(initial_vmap, mb_size=32, monitors=[m, mce, free_energy], name='train', mode=mode)

morb/base.py

@@ -44,7 +44,7 @@ def mean_field(self, vmap, pmap):
         return self.mean_field_from_activation({ self: self.activation(vmap, pmap) })
 
     def free_energy_term(self, vmap, pmap):
-        return self.free_energy_term_from_activation({ self: self.activation(vmap, pmap) })
+        return self.free_energy_term_from_activation({ self: self.activation(vmap, pmap) }, pmap)
 
     def log_prob(self, vmap, pmap):
         activation_vmap = { self: self.activation(vmap, pmap) }

morb/rbms.py

@@ -19,9 +19,14 @@ def __init__(self, n_visible, n_hidden):
         self.v = units.BinaryUnits(self, name='v') # visibles
         self.h = units.BinaryUnits(self, name='h') # hiddens
         # parameters
-        self.W = parameters.ProdParameters(self, [self.v, self.h], theano.shared(value = self._initial_W(), name='W'), name='W') # weights
-        self.bv = parameters.BiasParameters(self, self.v, theano.shared(value = self._initial_bv(), name='bv'), name='bv') # visible bias
-        self.bh = parameters.BiasParameters(self, self.h, theano.shared(value = self._initial_bh(), name='bh'), name='bh') # hidden bias
+        self.pmap = {
+            'W': theano.shared(value = self._initial_W(), name='W'),
+            'bv': theano.shared(value = self._initial_bv(), name='bv'),
+            'bh': theano.shared(value = self._initial_bh(), name='bh')
+          }
+        self.W = parameters.ProdParameters(self, [self.v, self.h], 'W', name='W') # weights
+        self.bv = parameters.BiasParameters(self, self.v, 'bv', name='bv') # visible bias
+        self.bh = parameters.BiasParameters(self, self.h, 'bh', name='bh') # hidden bias
 
     def _initial_W(self):
         return np.asarray( np.random.uniform(
@@ -46,8 +51,10 @@ def __init__(self, n_visible, n_hidden, n_context):
         # units
         self.x = units.Units(self, name='x') # context
         # parameters
-        self.A = parameters.ProdParameters(self, [self.x, self.v], theano.shared(value = self._initial_A(), name='A'), name='A') # context-to-visible weights
-        self.B = parameters.ProdParameters(self, [self.x, self.h], theano.shared(value = self._initial_B(), name='B'), name='B') # context-to-hidden weights
+        self.pmap['A'] = theano.shared(value = self._initial_A(), name='A')
+        self.pmap['B'] = theano.shared(value = self._initial_B(), name='B')
+        self.A = parameters.ProdParameters(self, [self.x, self.v], 'A', name='A') # context-to-visible weights
+        self.B = parameters.ProdParameters(self, [self.x, self.h], 'B', name='B') # context-to-hidden weights
 
     def _initial_A(self):
         return np.zeros((self.n_context, self.n_visible), dtype = theano.config.floatX)