Do not add internal RandomVariable's in RV collection

edward/inferences/conjugacy/conjugacy.py

@@ -71,7 +71,11 @@ def complete_conditional(rv, cond_set=None):
   result in unpredictable behavior.
   """
   if cond_set is None:
+    # Default to Markov blanket, excluding conditionals. This is useful if
+    # calling complete_conditional many times without passing in cond_set.
     cond_set = get_blanket(rv)
+    cond_set = [i for i in cond_set if not
+                ('complete_conditional' in i.name and 'cond_dist' in i.name)]
 
   cond_set = set([rv] + list(cond_set))
   with tf.name_scope('complete_conditional_%s' % rv.name) as scope:
@@ -143,7 +147,9 @@ def complete_conditional(rv, cond_set=None):
 
 def get_log_joint(cond_set):
   g = tf.get_default_graph()
-  cond_set_name = 'log_joint_of_' + ('_'.join([i.name[:-1] for i in cond_set]))
+  cond_set_names = [i.name[:-1] for i in cond_set]
+  cond_set_names.sort()
+  cond_set_name = 'log_joint_of_' + '_'.join(cond_set_names)
   with tf.name_scope("conjugate_log_joint/") as scope:
     try:
       # Use log joint tensor if already built in graph.

edward/models/dirichlet_process.py

@@ -64,7 +64,8 @@ def __init__(self, alpha, base, validate_args=False, allow_nan_stats=True,
             dtype=self._base.dtype)
 
         # Instantiate distribution for stick breaking proportions.
-        self._betadist = Beta(a=tf.ones_like(self._alpha), b=self._alpha)
+        self._betadist = Beta(a=tf.ones_like(self._alpha), b=self._alpha,
+                              collections=[])
         # Form empty tensor to store stick breaking proportions.
         self._beta = tf.zeros(
             [0] + self.get_batch_shape().as_list(),

edward/models/param_mixture.py

@@ -82,6 +82,7 @@ def __init__(self,
       self._components = component_dist(validate_args=validate_args,
                                         allow_nan_stats=allow_nan_stats,
                                         sample_shape=sample_shape,
+                                        collections=[],
                                         **component_params)
 
       if validate_args:

edward/models/random_variable.py

@@ -10,7 +10,7 @@
 except Exception as e:
   raise ImportError("{0}. Your TensorFlow version is not supported.".format(e))
 
-RANDOM_VARIABLE_COLLECTION = "_random_variable_collection_"
+RANDOM_VARIABLE_COLLECTION = "random_variables"
 
 
 class RandomVariable(object):
@@ -78,6 +78,9 @@ def __init__(self, *args, **kwargs):
     value : tf.Tensor, optional
       Fixed tensor to associate with random variable. Must have shape
       ``sample_shape + batch_shape + event_shape``.
+    collections : list, optional
+      Optional list of graph collections keys. The random variable is
+      added to these collections. Defaults to ["random_variables"].
     *args, **kwargs
       Passed into parent ``__init__``.
     """
@@ -88,11 +91,14 @@ def __init__(self, *args, **kwargs):
     # temporarily pop (then reinsert) before calling parent __init__
     sample_shape = kwargs.pop('sample_shape', ())
     value = kwargs.pop('value', None)
+    collections = kwargs.pop('collections', [RANDOM_VARIABLE_COLLECTION])
     super(RandomVariable, self).__init__(*args, **kwargs)
     if sample_shape != ():
       self._kwargs['sample_shape'] = sample_shape
     if value is not None:
       self._kwargs['value'] = value
+    if collections != [RANDOM_VARIABLE_COLLECTION]:
+      self._kwargs['collections'] = collections
 
     self._sample_shape = tf.TensorShape(sample_shape)
     if value is not None:
@@ -115,7 +121,8 @@ def __init__(self, *args, **kwargs):
             "value argument or implement sample for {0}."
             .format(self.__class__.__name__))
 
-    tf.add_to_collection(RANDOM_VARIABLE_COLLECTION, self)
+    for collection in collections:
+      tf.add_to_collection(collection, self)
 
   @property
   def shape(self):

examples/mixture_gaussian_gibbs.py

@@ -41,11 +41,10 @@
 z = x.cat
 
 # Conditionals
-blanket = [x, z, mu, pi, sigmasq]
-mu_cond = ed.complete_conditional(mu, blanket)
-sigmasq_cond = ed.complete_conditional(sigmasq, blanket)
-pi_cond = ed.complete_conditional(pi, blanket)
-z_cond = ed.complete_conditional(z, blanket)
+mu_cond = ed.complete_conditional(mu)
+sigmasq_cond = ed.complete_conditional(sigmasq)
+pi_cond = ed.complete_conditional(pi)
+z_cond = ed.complete_conditional(z)
 
 sess = ed.get_session()
 

tests/test-inferences/test_conjugacy.py

@@ -180,8 +180,7 @@ def test_dirichlet_categorical(self):
     theta = rvs.Dirichlet(alpha)
     x = rvs.Categorical(p=theta, sample_shape=sample_shape)
 
-    blanket = [theta, x]
-    theta_cond = ed.complete_conditional(theta, blanket)
+    theta_cond = ed.complete_conditional(theta, [theta, x])
 
     with self.test_session() as sess:
       alpha_val = sess.run(theta_cond.alpha, {x: x_data})
@@ -208,10 +207,9 @@ def test_mog(self):
                          rvs.Normal, sample_shape=N)
     z = x.cat
 
-    blanket = [x, z, mu, pi]
-    mu_cond = ed.complete_conditional(mu, blanket)
-    pi_cond = ed.complete_conditional(pi, blanket)
-    z_cond = ed.complete_conditional(z, blanket)
+    mu_cond = ed.complete_conditional(mu)
+    pi_cond = ed.complete_conditional(pi)
+    z_cond = ed.complete_conditional(z)
 
     with self.test_session() as sess:
       pi_cond_alpha, mu_cond_mu, mu_cond_sigma, z_cond_p = (