add names to all tf.variables internal to inference

edward/inferences/gan_inference.py

@@ -201,7 +201,7 @@ def print_progress(self, info_dict):
 def _build_optimizer(optimizer, global_step):
   if optimizer is None:
     # Use ADAM with a decaying scale factor.
-    global_step = tf.Variable(0, trainable=False)
+    global_step = tf.Variable(0, trainable=False, name="global_step")
     starter_learning_rate = 0.1
     learning_rate = tf.train.exponential_decay(starter_learning_rate,
                                                global_step,

edward/inferences/inference.py

@@ -62,10 +62,11 @@ def __init__(self, latent_vars=None, data=None):
         elif isinstance(value, (float, list, int, np.ndarray, np.number, str)):
           # If value is a Python type, store it in the graph.
           # Assign its placeholder with the key's data type.
-          ph = tf.placeholder(key.dtype, np.shape(value))
-          var = tf.Variable(ph, trainable=False, collections=[])
-          sess.run(var.initializer, {ph: value})
-          self.data[key] = var
+          with tf.variable_scope("data"):
+            ph = tf.placeholder(key.dtype, np.shape(value))
+            var = tf.Variable(ph, trainable=False, collections=[])
+            sess.run(var.initializer, {ph: value})
+            self.data[key] = var
 
   def run(self, variables=None, use_coordinator=True, *args, **kwargs):
     """A simple wrapper to run inference.
@@ -163,7 +164,8 @@ def initialize(self, n_iter=1000, n_print=None, scale=None, logdir=None,
       self.n_print = n_print
 
     self.progbar = Progbar(self.n_iter)
-    self.t = tf.Variable(0, trainable=False)
+    self.t = tf.Variable(0, trainable=False, name="iteration")
+
     self.increment_t = self.t.assign_add(1)
 
     if scale is None:

edward/inferences/monte_carlo.py

@@ -83,7 +83,7 @@ def initialize(self, *args, **kwargs):
                                 qz in six.itervalues(self.latent_vars)])
     super(MonteCarlo, self).initialize(*args, **kwargs)
 
-    self.n_accept = tf.Variable(0, trainable=False)
+    self.n_accept = tf.Variable(0, trainable=False, name="n_accept")
     self.n_accept_over_t = self.n_accept / self.t
     self.train = self.build_update()
 

edward/inferences/variational_inference.py

@@ -65,7 +65,7 @@ def initialize(self, optimizer=None, var_list=None, use_prettytensor=False,
 
     if optimizer is None:
       # Use ADAM with a decaying scale factor.
-      global_step = tf.Variable(0, trainable=False)
+      global_step = tf.Variable(0, trainable=False, name="global_step")
       starter_learning_rate = 0.1
       learning_rate = tf.train.exponential_decay(starter_learning_rate,
                                                  global_step,