Change apply_column_names default to false (#56)

* Change apply_column_names default to false

* Add deprecation message

* flake8 fix

* More flake8

* Update tests. Update docs for pandas parameter type.

* Use a different name to avoid conflicts in the global dict

* Missed changing the return

Author: trangevi

inference_schema/parameter_types/pandas_parameter_type.py

@@ -7,17 +7,23 @@
 from .abstract_parameter_type import AbstractParameterType
 from ._util import get_swagger_for_list, get_swagger_for_nested_dict
 from ._constants import SWAGGER_FORMAT_CONSTANTS
+from warnings import warn
 
 
 class PandasParameterType(AbstractParameterType):
     """
     Class used to specify an expected parameter as a Pandas type.
     """
 
-    def __init__(self, sample_input, enforce_column_type=True, enforce_shape=True, apply_column_names=True,
+    def __init__(self, sample_input, enforce_column_type=True, enforce_shape=True, apply_column_names=False,
                  orient='records'):
         """
-        Construct the PandasParameterType object.
+        Construct the PandasParameterType object. An important note regarding Pandas DataFrame handling; by default,
+        Pandas supports integer type column names in a DataFrame. However, when using the built in methods for
+        converting a json object to a DataFrame, unless all of the columns are integers, they will all be converted
+        to strings. This ParameterType uses the built in methods for performing conversion, and as such
+        `deserialize_input` has the same limitation. It is recommended to not use a mix of string and integer
+        type column names in the provided sample, as this can lead to inconsistent/unexpected behavior.
 
         :param sample_input: A sample input dataframe. This sample will be used as a basis for column types and array
             shape.
@@ -28,8 +34,10 @@ def __init__(self, sample_input, enforce_column_type=True, enforce_shape=True, a
         :param enforce_shape: Enforce that input shape must match that of the provided sample when `deserialize_input`
             is called.
         :type enforce_shape: bool
-        :param apply_column_names: Apply column names from the provided sample onto the input when `deserialize_input`
-            is called.
+        :param apply_column_names: [DEPRECATED] Apply column names from the provided sample onto the input when
+            `deserialize_input` is called. Disabled by default, as there is no guaranteed order for dictionary keys,
+            so it's possible for names to be applied in the wrong order when `deserialize_input` is called. The
+            property is deprecated, and will be removed in a future update.
         :type apply_column_names: bool
         :param orient: The Pandas orient to use when converting between a json object and a DataFrame. Possible orients
             are 'split', 'records', 'index', 'columns', 'values', or 'table'. More information about these orients can
@@ -42,6 +50,10 @@ def __init__(self, sample_input, enforce_column_type=True, enforce_shape=True, a
         super(PandasParameterType, self).__init__(sample_input)
         self.enforce_column_type = enforce_column_type
         self.enforce_shape = enforce_shape
+
+        if apply_column_names:
+            warn('apply_column_names is a deprecated parameter and will be removed in a future update',
+                 DeprecationWarning, stacklevel=2)
         self.apply_column_names = apply_column_names
 
         if orient not in ('split', 'records', 'index', 'columns', 'values', 'table'):

tests/conftest.py

@@ -58,8 +58,8 @@ def pandas_sample_output():
 
 
 @pytest.fixture(scope="session")
-def pandas_sample_input_multi_type_column_labels():
-    pandas_input_data = {'name': ['Sarah', 'John'], 1: ['WA', 'CA']}
+def pandas_sample_input_int_column_labels():
+    pandas_input_data = {0: ['Sarah', 'John'], 1: ['WA', 'CA']}
     return pd.DataFrame(data=pandas_input_data)
 
 
@@ -121,22 +121,22 @@ def pandas_split_orient_func(param):
 
 
 @pytest.fixture(scope="session")
-def decorated_pandas_func_multi_type_column_labels(pandas_sample_input_multi_type_column_labels):
+def decorated_pandas_func_int_column_labels(pandas_sample_input_int_column_labels):
 
-    @input_schema('param', PandasParameterType(pandas_sample_input_multi_type_column_labels))
-    def pandas_split_orient_func(param):
+    @input_schema('param', PandasParameterType(pandas_sample_input_int_column_labels))
+    def pandas_int_column_labels_func(param):
         """
 
         :param param:
         :type param: pd.DataFrame
         :return:
         :rtype: pd.DataFrame
         """
-        assert param["name"] is not None
+        assert param[0] is not None
         assert param[1] is not None
         return param
 
-    return pandas_split_orient_func
+    return pandas_int_column_labels_func
 
 
 @pytest.fixture(scope="session")

tests/test_pandas_parameter_type.py

@@ -46,11 +46,11 @@ def test_pandas_timestamp_handling(self, decorated_pandas_datetime_func):
         result = decorated_pandas_datetime_func(**pandas_input)
         assert_frame_equal(result, datetime)
 
-    def test_pandas_multi_type_columns_labels_handling(self, decorated_pandas_func_multi_type_column_labels):
-        pandas_input = {'name': ['Sarah', 'John'], 1: ['WA', 'CA']}
-        result = decorated_pandas_func_multi_type_column_labels(pandas_input)
-        expected_result = pd.DataFrame(pandas_input)
-        assert_frame_equal(result, expected_result)
+    def test_pandas_int_column_labels(self, decorated_pandas_func_int_column_labels,
+                                      pandas_sample_input_int_column_labels):
+        input = pandas_sample_input_int_column_labels.to_dict(orient='records')
+        result = decorated_pandas_func_int_column_labels(input)
+        assert_frame_equal(result, pandas_sample_input_int_column_labels)
 
 
 class TestNestedType(object):