doc update and inputsdesc name check (#564)

.github/ISSUE_TEMPLATE.md

 Bug Reports/Feature Requests/Usage Questions Only:
 
-Bug reports or other problems with code: PLEASE always include
+Any unexpected problems with code: PLEASE always include
 1. What you did. (command you run and changes you made if using examples; post or describe your code if not)
 2. What you observed, e.g. as much as logs possible.
 3. What you expected, if not obvious.

docs/tutorial/symbolic.md

@@ -42,16 +42,21 @@ l = func(l, *args, **kwargs)
 l = FullyConnected('fc1', l, 10, nl=tf.identity)
 ```
 
-### Access Internal Variables:
+### Access Relevant Tensors
 
-Access the variables like this:
+The variables inside the layer will be named `name/W`, `name/b`, etc.
+See the API documentation of each layer for details.
+When building the graph, you can access the variables like this:
 ```python
 l = Conv2D('conv1', l, 32, 3)
 print(l.variables.W)
 print(l.variables.b)
 ```
-The names are documented in API documentation.
-Note that this method doesn't work with LinearWrap, and cannot access the variables created by an activation function.
+But note that this is a hacky way and may not work with future versions of TensorFlow.
+Also this method doesn't work with LinearWrap, and cannot access the variables created by an activation function.
+
+The output of a layer is usually named `name/output` unless documented differently in the API.
+You can always print a tensor to see its name.
 
 ### Use Models outside Tensorpack
 

tensorpack/graph_builder/model_desc.py

@@ -35,6 +35,8 @@ class InputDesc(
         """
         shape = tuple(shape)    # has to be tuple for "self" to be hashable
         assert isinstance(type, tf.DType), type
+        if any(k in name for k in [':', '/', ' ']):
+            raise ValueError("Invalid InputDesc name: '{}'".format(name))
         self = super(InputDesc, cls).__new__(cls, type, shape, name)
         self._cached_placeholder = None
         return self

tensorpack/tfutils/tower.py

@@ -170,12 +170,11 @@ def get_current_tower_context():
 
 class TowerFuncWrapper(object):
     """
-    A wrapper around a function which builds one tower (one replicate of the model).
+    A wrapper around a tower function (function which builds one tower, i.e. one replicate of the model).
     It keeps track of the name scope, variable scope and input/output tensors
     each time the function is called.
 
-    :class:`TowerTrainer` needs this option to be set, so that
-    it knows how to build a predictor.
+    :class:`TowerTrainer` needs this so that it knows how to build a predictor.
     """
 
     def __init__(self, tower_fn, inputs_desc):
@@ -186,11 +185,13 @@ class TowerFuncWrapper(object):
             inputs_desc ([InputDesc]): use this to figure out the right name for the input tensors.
         """
         assert callable(tower_fn), tower_fn
-        if not isinstance(tower_fn, TowerFuncWrapper):
-            self._tower_fn = tower_fn
-            self._inputs_desc = inputs_desc
+        inputs_desc_names = [k.name for k in inputs_desc]
+        assert len(set(inputs_desc_names)) == len(inputs_desc_names), \
+            "Duplicated names in inputs_desc! " + str(inputs_desc_names)
+        self._tower_fn = tower_fn
+        self._inputs_desc = inputs_desc
 
-            self._handles = []
+        self._handles = []
 
     def __new__(cls, tower_fn, inputs_desc):
         # to avoid double-wrapping a function