API: add infer_objects for soft conversions

doc/source/api.rst

@@ -270,6 +270,7 @@ Conversion
    :toctree: generated/
 
    Series.astype
+   Series.infer_objects
    Series.copy
    Series.isnull
    Series.notnull
@@ -777,6 +778,7 @@ Conversion
 
    DataFrame.astype
    DataFrame.convert_objects
+   DataFrame.infer_objects
    DataFrame.copy
    DataFrame.isnull
    DataFrame.notnull

doc/source/basics.rst

@@ -2024,7 +2024,22 @@ object conversion
 ~~~~~~~~~~~~~~~~~
 
 pandas offers various functions to try to force conversion of types from the ``object`` dtype to other types.
-The following functions are available for one dimensional object arrays or scalars:
+In cases where the data is already of the correct type, but stored in an ``object`` array, the
+:meth:`~DataFrame.infer_objects` and :meth:`~Series.infer_objects` can be used to soft convert
+to the correct type.
+
+  .. ipython:: python
+
+     df = pd.DataFrame([[1, 2],
+                        ['a', 'b'],
+                        [datetime.datetime(2016, 3, 2), datetime.datetime(2016, 3, 2)]])
+     df = df.T
+     df
+     df.dtypes
+     df.infer_objects().dtypes
+
+The following functions are available for one dimensional object arrays or scalars to perform
+hard conversion of objects to a specified type:
 
 - :meth:`~pandas.to_numeric` (conversion to numeric dtypes)
 

doc/source/whatsnew/v0.21.0.txt

@@ -25,6 +25,39 @@ New features
 - Added ``__fspath__`` method to :class:`~pandas.HDFStore`, :class:`~pandas.ExcelFile`,
   and :class:`~pandas.ExcelWriter` to work properly with the file system path protocol (:issue:`13823`)
 
+
+.. _whatsnew_0210.enhancements.infer_objects:
+
+``infer_objects`` type conversion
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+The `:meth:`~DataFrame.infer_objects` and :meth:`~Series.infer_objects`
+have been added  to perform dtype inference on object columns, replacing
+some of the functionality of the deprecated ``convert_objects``
+method (:issue:`11221`)
+
+This function only performs soft conversions on object columns, converting Python objects
+to native types, but not any coercive conversions.  For example:
+
+.. ipython:: python
+
+   df = pd.DataFrame({'A': [1, 2, 3],
+                      'B': np.array([1, 2, 3], dtype='object'),
+                      'C': ['1', '2', '3']})
+   df.dtypes
+   df.infer_objects().dtype
+
+Note that column ``'C'`` was not converted - only scalar numeric types
+will be inferred to a new type.  Other types of conversion should be accomplished
+using :func:`to_numeric` function (or :func:`to_datetime`, :func:`to_timedelta`).  See
+the documentation :ref:`here <basics.object_conversion>` for more details.
+
+.. ipython:: python
+
+   df = df.infer_objects()
+   df['C'] = pd.to_numeric(df['C'], errors='coerce')
+   df.dtypes
+
 .. _whatsnew_0210.enhancements.other:
 
 Other Enhancements

pandas/core/generic.py

@@ -3634,16 +3634,64 @@ def convert_objects(self, convert_dates=True, convert_numeric=False,
         converted : same as input object
         """
         from warnings import warn
-        warn("convert_objects is deprecated.  Use the data-type specific "
-             "converters pd.to_datetime, pd.to_timedelta and pd.to_numeric.",
-             FutureWarning, stacklevel=2)
+        msg = ("convert_objects is deprecated.  To re-infer data dtypes for "
+               "object columns, use {klass}.infer_objects()\nFor all "
+               "other conversions use the data-type specific converters "
+               "pd.to_datetime, pd.to_timedelta and pd.to_numeric."
+               ).format(klass=self.__class__.__name__)
+        warn(msg, FutureWarning, stacklevel=2)
 
         return self._constructor(
             self._data.convert(convert_dates=convert_dates,
                                convert_numeric=convert_numeric,
                                convert_timedeltas=convert_timedeltas,
                                copy=copy)).__finalize__(self)
 
+    def infer_objects(self):
+        """
+        Attempt to infer better dtypes for object columns.
+
+        Attempts soft conversion of object-dtyped
+        columns, leaving non-object and unconvertible
+        columns unchanged. The inference rules are the
+        same as during normal Series/DataFrame construction.
+
+        .. versionadded:: 0.20.0
+
+        See Also
+        --------
+        pandas.to_datetime : Convert argument to datetime.
+        pandas.to_timedelta : Convert argument to timedelta.
+        pandas.to_numeric : Convert argument to numeric typeR
+
+        Returns
+        -------
+        converted : same type as input object
+
+        Examples
+        --------
+        >>> df = pd.DataFrame({"A": ["a", 1, 2, 3]})
+        >>> df = df.iloc[1:]
+        >>> df
+           A
+        1  1
+        2  2
+        3  3
+        >>> df.dtypes
+        A    object
+        dtype: object
+        >>> df.infer_objects().dtypes
+        A    int64
+        dtype: object
+        """
+        # numeric=False necessary to only soft convert;
+        # python objects will still be converted to
+        # native numpy numeric types
+        return self._constructor(
+            self._data.convert(datetime=True, numeric=False,
+                               timedelta=True, coerce=False,
+                               copy=True)).__finalize__(self)
+
     # ----------------------------------------------------------------------
     # Filling NA's
 

pandas/tests/frame/test_block_internals.py

@@ -495,6 +495,32 @@ def test_convert_objects_no_conversion(self):
         mixed2 = mixed1._convert(datetime=True)
         assert_frame_equal(mixed1, mixed2)
 
+    def test_infer_objects(self):
+        # GH 11221
+        df = DataFrame({'a': ['a', 1, 2, 3],
+                        'b': ['b', 2.0, 3.0, 4.1],
+                        'c': ['c', datetime(2016, 1, 1),
+                              datetime(2016, 1, 2),
+                              datetime(2016, 1, 3)],
+                        'd': [1, 2, 3, 'd']},
+                       columns=['a', 'b', 'c', 'd'])
+        df = df.iloc[1:].infer_objects()
+
+        assert df['a'].dtype == 'int64'
+        assert df['b'].dtype == 'float64'
+        assert df['c'].dtype == 'M8[ns]'
+        assert df['d'].dtype == 'object'
+
+        expected = DataFrame({'a': [1, 2, 3],
+                              'b': [2.0, 3.0, 4.1],
+                              'c': [datetime(2016, 1, 1),
+                                    datetime(2016, 1, 2),
+                                    datetime(2016, 1, 3)],
+                              'd': [2, 3, 'd']},
+                             columns=['a', 'b', 'c', 'd'])
+        # reconstruct frame to verify inference is same
+        tm.assert_frame_equal(df.reset_index(drop=True), expected)
+
     def test_stale_cached_series_bug_473(self):
 
         # this is chained, but ok

pandas/tests/series/test_dtypes.py

@@ -268,3 +268,21 @@ def test_series_to_categorical(self):
         expected = Series(['a', 'b', 'c'], dtype='category')
 
         tm.assert_series_equal(result, expected)
+
+    def test_infer_objects_series(self):
+        # GH 11221
+        actual = Series(np.array([1, 2, 3], dtype='O')).infer_objects()
+        expected = Series([1, 2, 3])
+        tm.assert_series_equal(actual, expected)
+
+        actual = Series(np.array([1, 2, 3, None], dtype='O')).infer_objects()
+        expected = Series([1., 2., 3., np.nan])
+        tm.assert_series_equal(actual, expected)
+
+        # only soft conversions, uncovertable pass thru unchanged
+        actual = (Series(np.array([1, 2, 3, None, 'a'], dtype='O'))
+                  .infer_objects())
+        expected = Series([1, 2, 3, None, 'a'])
+
+        assert actual.dtype == 'object'
+        tm.assert_series_equal(actual, expected)