ENH: added support for Index.sort_values(key=...)

Author: jacobaustin123

pandas/core/frame.py

@@ -15,7 +15,7 @@
 import itertools
 import sys
 from textwrap import dedent
-from typing import FrozenSet, List, Optional, Set, Type, Union
+from typing import Callable, FrozenSet, List, Optional, Set, Type, Union
 import warnings
 
 import numpy as np
@@ -4977,7 +4977,7 @@ def sort_values(
         inplace=False,
         kind="quicksort",
         na_position="last",
-        key=None
+        key : Union[Callable, None] = None
     ):
         inplace = validate_bool_kwarg(inplace, "inplace")
         axis = self._get_axis_number(axis)
@@ -4992,29 +4992,20 @@ def sort_values(
         if len(by) > 1:
             from pandas.core.sorting import lexsort_indexer
 
-            if key is not None:
-                key_func = np.vectorize(key)
-                keys = [key_func(self._get_label_or_level_values(x, axis=axis)) for x in by]
-            else:
-                keys = [self._get_label_or_level_values(x, axis=axis) for x in by]
-
-            indexer = lexsort_indexer(keys, orders=ascending, na_position=na_position)
+            keys = [self._get_label_or_level_values(x, axis=axis) for x in by]
+            indexer = lexsort_indexer(keys, orders=ascending, na_position=na_position, key=key)
             indexer = ensure_platform_int(indexer)
         else:
             from pandas.core.sorting import nargsort
 
             by = by[0]
             k = self._get_label_or_level_values(by, axis=axis)
 
-            if key is not None:
-                key_func = np.vectorize(key)
-                k = key_func(k)
-
             if isinstance(ascending, (tuple, list)):
                 ascending = ascending[0]
 
             indexer = nargsort(
-                k, kind=kind, ascending=ascending, na_position=na_position
+                k, kind=kind, ascending=ascending, na_position=na_position, key=key
             )
 
         new_data = self._data.take(
@@ -5038,7 +5029,7 @@ def sort_index(
         na_position="last",
         sort_remaining=True,
         by=None,
-        key=None
+        key : Union[Callable, None] = None
     ):
 
         # TODO: this can be combined with Series.sort_index impl as
@@ -5060,7 +5051,7 @@ def sort_index(
         axis = self._get_axis_number(axis)
         labels = self._get_axis(axis)
         if key is not None:
-            labels = labels.map(key)
+            labels = labels.map(key, na_action="ignore")
 
         # make sure that the axis is lexsorted to start
         # if not we need to reconstruct to get the correct indexer

pandas/core/indexes/base.py

@@ -1,7 +1,7 @@
 from datetime import datetime, timedelta
 import operator
 from textwrap import dedent
-from typing import Union
+from typing import Callable, Union
 import warnings
 
 import numpy as np
@@ -4524,7 +4524,7 @@ def asof_locs(self, where, mask):
 
         return result
 
-    def sort_values(self, return_indexer=False, ascending=True):
+    def sort_values(self, return_indexer=False, ascending=True, key : Callable = None):
         """
         Return a sorted copy of the index.
 
@@ -4537,6 +4537,9 @@ def sort_values(self, return_indexer=False, ascending=True):
             Should the indices that would sort the index be returned.
         ascending : bool, default True
             Should the index values be sorted in an ascending order.
+        key : Callable, default None
+            Apply a key function to the indices before sorting, like 
+            built-in sorted function.
 
         Returns
         -------
@@ -4567,7 +4570,12 @@ def sort_values(self, return_indexer=False, ascending=True):
         >>> idx.sort_values(ascending=False, return_indexer=True)
         (Int64Index([1000, 100, 10, 1], dtype='int64'), array([3, 1, 0, 2]))
         """
-        _as = self.argsort()
+        if key:
+            idx = self.map(key, na_action="ignore")
+        else:
+            idx = self
+
+        _as = idx.argsort()
         if not ascending:
             _as = _as[::-1]
 
@@ -4679,9 +4687,12 @@ def argsort(self, *args, **kwargs):
         >>> idx[order]
         Index(['a', 'b', 'c', 'd'], dtype='object')
         """
+
         result = self.asi8
+
         if result is None:
             result = np.array(self)
+
         return result.argsort(*args, **kwargs)
 
     _index_shared_docs[

pandas/core/series.py

@@ -5,7 +5,7 @@
 from io import StringIO
 from shutil import get_terminal_size
 from textwrap import dedent
-from typing import Any, Callable
+from typing import Any, Callable, Union
 import warnings
 
 import numpy as np
@@ -3019,7 +3019,7 @@ def sort_values(
         inplace=False,
         kind="quicksort",
         na_position="last",
-        key=None
+        key: Callable = None
     ):
         """
         Sort by the values.
@@ -3042,7 +3042,7 @@ def sort_values(
         na_position : {'first' or 'last'}, default 'last'
             Argument 'first' puts NaNs at the beginning, 'last' puts NaNs at
             the end.
-        key : function, default None
+        key : Callable, default None
             If not None, apply the key function to every value before
             sorting. Identical to key argument in built-in sorted function.
 
@@ -3220,7 +3220,7 @@ def sort_index(
         kind="quicksort",
         na_position="last",
         sort_remaining=True,
-        key=None
+        key : Callable = None
     ):
         """
         Sort Series by index labels.
@@ -3249,7 +3249,7 @@ def sort_index(
         sort_remaining : bool, default True
             If True and sorting by level and index is multilevel, sort by other
             levels too (in order) after sorting by specified level.
-        key : function, default None
+        key : Callable, default None
             If not None, apply the key function to every index element before
             sorting. Identical to key argument in built-in sorted function.
 
@@ -3356,7 +3356,7 @@ def sort_index(
         index = self.index
         true_index = index
         if key is not None:
-            index = index.map(key)
+            index = index.map(key, na_action="ignore")
 
         if level is not None:
             new_index, indexer = index.sortlevel(

pandas/core/sorting.py

@@ -1,4 +1,6 @@
 """ miscellaneous sorting / groupby utilities """
+from typing import Callable, Union
+
 import numpy as np
 
 from pandas._libs import algos, hashtable, lib
@@ -188,7 +190,7 @@ def indexer_from_factorized(labels, shape, compress=True):
     return get_group_index_sorter(ids, ngroups)
 
 
-def lexsort_indexer(keys, orders=None, na_position="last"):
+def lexsort_indexer(keys, orders=None, na_position="last", key : Union[Callable, None] = None):
     from pandas.core.arrays import Categorical
 
     labels = []
@@ -198,6 +200,10 @@ def lexsort_indexer(keys, orders=None, na_position="last"):
     elif orders is None:
         orders = [True] * len(keys)
 
+    if key:
+        key_func = np.vectorize(key)
+        keys = [key_func(entry) if entry.size != 0 else entry for entry in keys]
+
     for key, order in zip(keys, orders):
 
         # we are already a Categorical
@@ -234,7 +240,7 @@ def lexsort_indexer(keys, orders=None, na_position="last"):
     return indexer_from_factorized(labels, shape)
 
 
-def nargsort(items, kind="quicksort", ascending=True, na_position="last"):
+def nargsort(items, kind="quicksort", ascending=True, na_position="last", key: Union[Callable, None] = None):
     """
     This is intended to be a drop-in replacement for np.argsort which
     handles NaNs. It adds ascending and na_position parameters.
@@ -250,6 +256,17 @@ def nargsort(items, kind="quicksort", ascending=True, na_position="last"):
     else:
         items = np.asanyarray(items)
 
+    if key is not None:
+        key_func = np.vectorize(key)
+        masked = np.ma.MaskedArray(items, mask)
+
+        if masked.size == 0:
+            vals = np.array([])  # vectorize fails on empty object arrays
+        else:
+            vals = np.asarray(key_func(masked))   # revert from masked
+
+        return nargsort(vals, kind=kind, ascending=ascending, na_position=na_position, key=None)
+
     idx = np.arange(len(items))
     non_nans = items[~mask]
     non_nan_idx = idx[~mask]

pandas/tests/frame/test_sorting.py

@@ -83,29 +83,33 @@ def test_sort_values(self):
         with pytest.raises(ValueError, match=msg):
             frame.sort_values(by=["A", "B"], axis=0, ascending=[True] * 5)
 
-    def test_sort_values_inplace(self):
+    @pytest.fixture(params=[None, lambda x : x])
+    def key(self, request):
+        return request.param
+
+    def test_sort_values_inplace(self, key):
         frame = DataFrame(
             np.random.randn(4, 4), index=[1, 2, 3, 4], columns=["A", "B", "C", "D"]
         )
 
         sorted_df = frame.copy()
-        sorted_df.sort_values(by="A", inplace=True)
-        expected = frame.sort_values(by="A")
+        sorted_df.sort_values(by="A", inplace=True, key=key)
+        expected = frame.sort_values(by="A", key=key)
         assert_frame_equal(sorted_df, expected)
 
         sorted_df = frame.copy()
-        sorted_df.sort_values(by=1, axis=1, inplace=True)
-        expected = frame.sort_values(by=1, axis=1)
+        sorted_df.sort_values(by=1, axis=1, inplace=True, key=key)
+        expected = frame.sort_values(by=1, axis=1, key=key)
         assert_frame_equal(sorted_df, expected)
 
         sorted_df = frame.copy()
-        sorted_df.sort_values(by="A", ascending=False, inplace=True)
-        expected = frame.sort_values(by="A", ascending=False)
+        sorted_df.sort_values(by="A", ascending=False, inplace=True, key=key)
+        expected = frame.sort_values(by="A", ascending=False, key=key)
         assert_frame_equal(sorted_df, expected)
 
         sorted_df = frame.copy()
-        sorted_df.sort_values(by=["A", "B"], ascending=False, inplace=True)
-        expected = frame.sort_values(by=["A", "B"], ascending=False)
+        sorted_df.sort_values(by=["A", "B"], ascending=False, inplace=True, key=key)
+        expected = frame.sort_values(by=["A", "B"], ascending=False, key=key)
         assert_frame_equal(sorted_df, expected)
 
     def test_sort_nan(self):
@@ -249,6 +253,23 @@ def test_sort_multi_index(self):
 
         tm.assert_frame_equal(result, expected)
 
+    def test_sort_multi_index_key(self):
+        # GH 25775, testing that sorting by index works with a multi-index.
+        df = DataFrame(
+            {"a": [3, 1, 2], "b": [0, 0, 0], "c": [0, 1, 2], "d": list("abc")}
+        )
+        result = df.set_index(list("abc")).sort_index(level=list("ba"), key=lambda x : x[0])
+
+        expected = DataFrame(
+            {"a": [1, 2, 3], "b": [0, 0, 0], "c": [1, 2, 0], "d": list("bca")}
+        )
+        expected = expected.set_index(list("abc"))
+        tm.assert_frame_equal(result, expected)
+
+        result = df.set_index(list("abc")).sort_index(level=list("ba"), key=lambda x : x[2])
+        expected = df.set_index(list("abc"))
+        tm.assert_frame_equal(result, expected)
+
     def test_stable_categorial(self):
         # GH 16793
         df = DataFrame({"x": pd.Categorical(np.repeat([1, 2, 3, 4], 5), ordered=True)})
@@ -628,6 +649,13 @@ def test_sort_value_key_nan(self):
         expected = df.sort_values(1, key=str.lower, ascending=False)
         assert_frame_equal(result, expected)
 
+    @pytest.mark.parametrize('key', [None, lambda x : x])
+    def test_sort_value_key_empty(self, key):
+        df = DataFrame(np.array([]))
+ 
+        df.sort_values(0, key=key)
+        df.sort_index(key=key)
+
     def test_sort_index(self):
         # GH13496
 

pandas/tests/indexing/multiindex/test_sorted.py

@@ -4,6 +4,8 @@
 from pandas import DataFrame, MultiIndex, Series
 from pandas.util import testing as tm
 
+import pytest
+
 
 class TestMultiIndexSorted:
     def test_getitem_multilevel_index_tuple_not_sorted(self):
@@ -28,7 +30,8 @@ def test_getitem_slice_not_sorted(self, multiindex_dataframe_random_data):
         expected = df.reindex(columns=df.columns[:3])
         tm.assert_frame_equal(result, expected)
 
-    def test_frame_getitem_not_sorted2(self):
+    @pytest.mark.parametrize('key', [None, lambda x : x])
+    def test_frame_getitem_not_sorted2(self, key):
         # 13431
         df = DataFrame(
             {
@@ -47,15 +50,35 @@ def test_frame_getitem_not_sorted2(self):
         assert not df2.index.is_monotonic
 
         assert df2_original.index.equals(df2.index)
-        expected = df2.sort_index()
+        expected = df2.sort_index(key=key)
         assert expected.index.is_lexsorted()
         assert expected.index.is_monotonic
 
-        result = df2.sort_index(level=0)
+        result = df2.sort_index(level=0, key=key)
         assert result.index.is_lexsorted()
         assert result.index.is_monotonic
         tm.assert_frame_equal(result, expected)
 
+    def test_sort_values_key(self, multiindex_dataframe_random_data):
+        arrays = [
+            ["bar", "bar", "baz", "baz", "qux", "qux", "foo", "foo"],
+            ["one", "two", "one", "two", "one", "two", "one", "two"],
+        ]
+        tuples = zip(*arrays)
+        index = MultiIndex.from_tuples(tuples)
+        index = index.sort_values(key=lambda x: (x[0][2], x[1][2]))
+        result = DataFrame(range(8), index=index)
+
+        arrays = [
+            ["foo", "foo", "bar", "bar", "qux", "qux", "baz", "baz"],
+            ["one", "two", "one", "two", "one", "two", "one", "two"],
+        ]
+        tuples = zip(*arrays)
+        index = MultiIndex.from_tuples(tuples)
+        expected = DataFrame(range(8), index=index)
+
+        tm.assert_frame_equal(result, expected)
+
     def test_frame_getitem_not_sorted(self, multiindex_dataframe_random_data):
         frame = multiindex_dataframe_random_data
         df = frame.T

pandas/tests/series/test_sorting.py

@@ -152,6 +152,18 @@ def test_sort_index_multiindex(self, level):
         res = s.sort_index(level=level, sort_remaining=False)
         assert_series_equal(s, res)
 
+    def test_sort_index_multiindex_key(self):
+
+        mi = MultiIndex.from_tuples([[1, 1, 3], [1, 1, 1]], names=list("ABC"))
+        s = Series([1, 2], mi)
+        backwards = s.iloc[[1, 0]]
+
+        res = s.sort_index(key=lambda x : x[2])
+        assert_series_equal(backwards, res)
+
+        res = s.sort_index(key=lambda x : x[1])   # nothing happens
+        assert_series_equal(s, res)
+
     def test_sort_index_kind(self):
         # GH #14444 & #13589:  Add support for sort algo choosing
         series = Series(index=[3, 2, 1, 4, 3])