diff -ur patsy-0.5.1/doc/stateful-transforms.rst patsy-0.5.1.doc/doc/stateful-transforms.rst
--- patsy-0.5.1/doc/stateful-transforms.rst	2018-10-28 04:04:50.000000000 +0100
+++ patsy-0.5.1.doc/doc/stateful-transforms.rst	2018-11-12 21:25:46.469485037 +0100
@@ -35,6 +35,7 @@
 Now we can build a design matrix and see what we get:
 
 .. ipython:: python
+   :okexcept:
 
    mat = dmatrix("naive_center(x)", data)
    mat
@@ -49,6 +50,7 @@
 transformation, like so:
 
 .. ipython:: python
+   :okexcept:
 
    new_data = {"x": [5, 6, 7, 8]}
    # Broken!
@@ -76,6 +78,7 @@
 way:
 
 .. ipython:: python
+   :okexcept:
 
    fixed_mat = dmatrix("center(x)", data)
    fixed_mat
@@ -83,6 +86,7 @@
 But if we then feed in our new data, we also get out the correct result:
 
 .. ipython:: python
+   :okexcept:
 
    # Correct!
    build_design_matrices([fixed_mat.design_info], new_data)[0]
@@ -97,6 +101,7 @@
 just similar enough for you to miss the problem until it's too late.)
 
 .. ipython:: python
+   :okexcept:
 
    data_chunked = [{"x": data["x"][:2]},
                    {"x": data["x"][2:]}]
@@ -108,6 +113,7 @@
 But if we use the proper stateful transform, this just works:
 
 .. ipython:: python
+   :okexcept:
 
    dinfo = incr_dbuilder("center(x)", lambda: iter(data_chunked))
    # Correct!
@@ -133,6 +139,7 @@
 for prediction as well:
 
 .. ipython:: python
+   :okexcept:
 
    # Correct!
    build_design_matrices([dinfo], new_data)[0]