Add example of how to preserve order of columns with usecols.

Author: eric

doc/source/io.rst

@@ -136,7 +136,11 @@ usecols : array-like or callable, default ``None``
   that correspond to column names provided either by the user in `names` or
   inferred from the document header row(s). For example, a valid array-like
   `usecols` parameter would be ``[0, 1, 2]`` or ``['foo', 'bar', 'baz']``. 
-  Element order is ignored, so ``usecols=[0, 1]`` is the same as ``[1, 0]``.
+  Element order is ignored, so ``usecols=[0, 1]`` is the same as ``[1, 0]``. Element
+  order is ignored, so usecols=[1,0] is the same as [0,1]. To instantiate a
+  DataFrame with element order preserved use ``pd.read_csv(usecols=[0, 1])[[0, 1]]``
+  for columns in ``[1, 0]`` order or ``pd.read_csv(usecols=[0, 1])[[0, 1]]``
+  for ``[0, 1]`` order.
 
   If callable, the callable function will be evaluated against the column names,
   returning names where the callable function evaluates to True:

pandas/io/parsers.py

@@ -102,7 +102,10 @@
     that correspond to column names provided either by the user in `names` or
     inferred from the document header row(s). For example, a valid array-like
     `usecols` parameter would be [0, 1, 2] or ['foo', 'bar', 'baz']. Element
-    order is ignored, so usecols=[1,0] is the same as [0,1].
+    order is ignored, so usecols=[1,0] is the same as [0,1]. To instantiate a
+    DataFrame with element order preserved use
+    ``pd.read_csv(usecols=[0, 1])[[0, 1]]`` for columns in ``[1, 0]``
+    order or ``pd.read_csv(usecols=[0, 1])[[0, 1]]`` for ``[0, 1]`` order.
 
     If callable, the callable function will be evaluated against the column
     names, returning names where the callable function evaluates to True. An