fix nested argscope (fix #1252)

docs/tutorial/philosophy/dataflow.md

@@ -161,6 +161,7 @@ or when you need to filter your data on the fly.
    but inefficient for generic data type or numpy arrays.
    Also, its implementation [does not always clean up the subprocesses correctly](https://github.com/pytorch/pytorch/issues/16608).
 
+Pytorch starts to improve on these bad assumptions (e.g., with [IterableDataset](https://github.com/pytorch/pytorch/pull/19228)).
 On the other hand, DataFlow:
 
 1. Is a pure iterator, not necessarily has a length or can be indexed. This is more generic.

tensorpack/tfutils/argscope.py

@@ -49,7 +49,8 @@ def argscope(layers, **kwargs):
         assert hasattr(l, 'symbolic_function'), "{} is not a registered layer".format(l.__name__)
         # _check_args_exist(l.symbolic_function)
 
-    new_scope = copy.copy(get_arg_scope())
+    # need to deepcopy so that changes to new_scope does not affect outer scope
+    new_scope = copy.deepcopy(get_arg_scope())
     for l in layers:
         new_scope[l.__name__].update(kwargs)
     _ArgScopeStack.append(new_scope)