Merge 4db1296 into 20749a1

patsy/categorical.py

@@ -42,6 +42,7 @@
                         have_pandas, have_pandas_categorical,
                         have_pandas_categorical_dtype,
                         safe_is_pandas_categorical,
+                        safe_pandas_Categorical_reorder,
                         pandas_Categorical_from_codes,
                         pandas_Categorical_categories,
                         pandas_Categorical_codes,
@@ -309,21 +310,26 @@ def categorical_to_int(data, levels, NA_action, origin=None):
     assert isinstance(levels, tuple)
     # In this function, missing values are always mapped to -1
 
+    if isinstance(data, _CategoricalBox):
+        if data.levels is not None and tuple(data.levels) != levels:
+            raise PatsyError("mismatching levels: expected %r, got %r"
+                             % (levels, tuple(data.levels)), origin)
+        data = data.data
+
     if safe_is_pandas_categorical(data):
         data_levels_tuple = tuple(pandas_Categorical_categories(data))
-        if not data_levels_tuple == levels:
+        if not set(data_levels_tuple) == set(levels):
             raise PatsyError("mismatching levels: expected %r, got %r"
                              % (levels, data_levels_tuple), origin)
+        if not data_levels_tuple == levels:
+            if isinstance(data, pandas.Categorical):
+                data = safe_pandas_Categorical_reorder(data, levels)
+            else:
+                data = safe_pandas_Categorical_reorder(data.cat, levels)
         # pandas.Categorical also uses -1 to indicate NA, and we don't try to
         # second-guess its NA detection, so we can just pass it back.
         return pandas_Categorical_codes(data)
 
-    if isinstance(data, _CategoricalBox):
-        if data.levels is not None and tuple(data.levels) != levels:
-            raise PatsyError("mismatching levels: expected %r, got %r"
-                             % (levels, tuple(data.levels)), origin)
-        data = data.data
-
     data = _categorical_shape_fix(data)
 
     try:
@@ -402,11 +408,21 @@ def Series_from_codes(codes, categories):
                           con([1, 0], ("a", "b")),
                           ("a", "c"),
                           NAAction())
+
+            # I don't think this test is necesssary. If user specifies
+            # specifies a custom order of the levels, shouldn't we allow
+            # them to be re-ordered on-the-fly? 
+
+            # Contradicts test_highlevel.test_C_and_pandas_categorical
+            # test where levels can be re-ordered.
+
+            """
             assert_raises(PatsyError,
                           categorical_to_int,
                           con([1, 0], ("a", "b")),
                           ("b", "a"),
                           NAAction())
+            """
 
     def t(data, levels, expected, NA_action=NAAction()):
         got = categorical_to_int(data, levels, NA_action)

patsy/util.py

@@ -21,6 +21,7 @@
            "no_pickling",
            "assert_no_pickling",
            "safe_string_eq",
+           "test_safe_pandas_Categorical_reorder",
            ]
 
 import sys
@@ -610,6 +611,27 @@ def test_pandas_Categorical_from_codes():
     assert np.all(np.asarray(c)[:-1] == ["b", "b", "a"])
     assert np.isnan(np.asarray(c)[-1])
 
+
+def safe_pandas_Categorical_reorder(categorical, newlevels):
+    if hasattr(categorical, 'reorder_categories'):
+        return categorical.reorder_categories(newlevels, ordered=False)
+    data = np.asarray(categorical).tolist()
+    data = [np.where(d==np.array(newlevels))[0][0]
+            if not pandas.isnull(d) else -1 for d in data]
+    return pandas_Categorical_from_codes(data, newlevels)
+
+
+def test_safe_pandas_Categorical_reorder():
+    c = pandas_Categorical_from_codes([1, 1, 0, -1], ["a", "b"])
+    c = safe_pandas_Categorical_reorder(c, ['b', 'a'])
+    assert np.all(np.asarray(c)[:-1] == ["b", "b", "a"])
+    assert np.isnan(np.asarray(c)[-1])
+    if hasattr(c, 'categories'):
+        assert np.all(c.categories==['b', 'a'])
+    else:
+        assert np.all(c.levels==['b', 'a'])
+
+
 # Needed to support pandas < 0.15
 def pandas_Categorical_categories(cat):
     # In 0.15+, a categorical Series has a .cat attribute which is similar to