update docs

docs/tutorial/intro.rst

@@ -15,7 +15,8 @@ The Tensorpack API brings speed and flexibility together.
 
 
 Is TensorFlow Slow?
-~~~~~~~~~~~~~~~~~~~~~
+*******************
+
 
 There is a common misconception,
 but no, it's not slow. But it's not easy to write it in an efficient way.

docs/tutorial/trainer.md

@@ -55,7 +55,7 @@ The tower function needs to follow some rules:
 
 	 Tensorpack layers create variables based on the name given to the layer:
 	 e.g., `Conv2D('test', x)` will open a variable scope named "test".
-     In order to respect the above rules, 
+     In order to respect the above rules,
 	 the name of the layer must not depend on name_scope's name or any tensor's name.
 4. It must __respect variable scope reuse__:
    * The creation of any trainable variables must __respect reuse__ variable scope.

tensorpack/utils/utils.py

@@ -69,17 +69,30 @@ def humanize_time_delta(sec):
 def change_env(name, val):
     """
     Args:
-        name(str), val(str):
+        name(str): name of the env var
+        val(str or None): the value, or set to None to clear the env var.
 
     Returns:
         a context where the environment variable ``name`` being set to
         ``val``. It will be set back after the context exits.
     """
     oldval = os.environ.get(name, None)
-    os.environ[name] = val
+
+    if val is None:
+        try:
+            del os.environ[name]
+        except KeyError:
+            pass
+    else:
+        os.environ[name] = val
+
     yield
+
     if oldval is None:
-        del os.environ[name]
+        try:
+            del os.environ[name]
+        except KeyError:
+            pass
     else:
         os.environ[name] = oldval