Bug: Save original index and remap after function completes

doc/source/whatsnew/v3.0.0.rst

@@ -61,6 +61,7 @@ Other enhancements
 - :meth:`Series.cummin` and :meth:`Series.cummax` now supports :class:`CategoricalDtype` (:issue:`52335`)
 - :meth:`Series.plot` now correctly handle the ``ylabel`` parameter for pie charts, allowing for explicit control over the y-axis label (:issue:`58239`)
 - :meth:`DataFrame.plot.scatter` argument ``c`` now accepts a column of strings, where rows with the same string are colored identically (:issue:`16827` and :issue:`16485`)
+- :meth:`Series.nlargest` uses a 'stable' sort internally and will preserve original ordering.
 - :class:`ArrowDtype` now supports ``pyarrow.JsonType`` (:issue:`60958`)
 - :class:`DataFrameGroupBy` and :class:`SeriesGroupBy` methods ``sum``, ``mean``, ``median``, ``prod``, ``min``, ``max``, ``std``, ``var`` and ``sem`` now accept ``skipna`` parameter (:issue:`15675`)
 - :class:`Rolling` and :class:`Expanding` now support ``nunique`` (:issue:`26958`)
@@ -591,6 +592,7 @@ Performance improvements
 - :func:`concat` returns a :class:`RangeIndex` column when possible when ``objs`` contains :class:`Series` and :class:`DataFrame` and ``axis=0`` (:issue:`58119`)
 - :func:`concat` returns a :class:`RangeIndex` level in the :class:`MultiIndex` result when ``keys`` is a ``range`` or :class:`RangeIndex` (:issue:`57542`)
 - :meth:`RangeIndex.append` returns a :class:`RangeIndex` instead of a :class:`Index` when appending values that could continue the :class:`RangeIndex` (:issue:`57467`)
+- :meth:`Series.nlargest` has improved performance when there are duplicate values in the index (:issue:`55767`)
 - :meth:`Series.str.extract` returns a :class:`RangeIndex` columns instead of an :class:`Index` column when possible (:issue:`57542`)
 - :meth:`Series.str.partition` with :class:`ArrowDtype` returns a :class:`RangeIndex` columns instead of an :class:`Index` column when possible (:issue:`57768`)
 - Performance improvement in :class:`DataFrame` when ``data`` is a ``dict`` and ``columns`` is specified (:issue:`24368`)

pandas/core/methods/selectn.py

@@ -11,6 +11,7 @@
 from typing import (
     TYPE_CHECKING,
     Generic,
+    Literal,
     cast,
     final,
 )
@@ -54,7 +55,9 @@
 
 
 class SelectN(Generic[NDFrameT]):
-    def __init__(self, obj: NDFrameT, n: int, keep: str) -> None:
+    def __init__(
+        self, obj: NDFrameT, n: int, keep: Literal["first", "last", "all"]
+    ) -> None:
         self.obj = obj
         self.n = n
         self.keep = keep
@@ -111,15 +114,25 @@ def compute(self, method: str) -> Series:
         if n <= 0:
             return self.obj[[]]
 
-        dropped = self.obj.dropna()
-        nan_index = self.obj.drop(dropped.index)
+        # Save index and reset to default index to avoid performance impact
+        # from when index contains duplicates
+        original_index: Index = self.obj.index
+        default_index = self.obj.reset_index(drop=True)
 
-        # slow method
-        if n >= len(self.obj):
+        # Slower method used when taking the full length of the series
+        # In this case, it is equivalent to a sort.
+        if n >= len(default_index):
             ascending = method == "nsmallest"
-            return self.obj.sort_values(ascending=ascending).head(n)
+            result = default_index.sort_values(ascending=ascending, kind="stable").head(
+                n
+            )
+            result.index = original_index.take(result.index)
+            return result
+
+        # Fast method used in the general case
+        dropped = default_index.dropna()
+        nan_index = default_index.drop(dropped.index)
 
-        # fast method
         new_dtype = dropped.dtype
 
         # Similar to algorithms._ensure_data
@@ -158,7 +171,7 @@ def compute(self, method: str) -> Series:
         else:
             kth_val = np.nan
         (ns,) = np.nonzero(arr <= kth_val)
-        inds = ns[arr[ns].argsort(kind="mergesort")]
+        inds = ns[arr[ns].argsort(kind="stable")]
 
         if self.keep != "all":
             inds = inds[:n]
@@ -173,7 +186,9 @@ def compute(self, method: str) -> Series:
             # reverse indices
             inds = narr - 1 - inds
 
-        return concat([dropped.iloc[inds], nan_index]).iloc[:findex]
+        result = concat([dropped.iloc[inds], nan_index]).iloc[:findex]
+        result.index = original_index.take(result.index)
+        return result
 
 
 class SelectNFrame(SelectN[DataFrame]):
@@ -192,7 +207,13 @@ class SelectNFrame(SelectN[DataFrame]):
     nordered : DataFrame
     """
 
-    def __init__(self, obj: DataFrame, n: int, keep: str, columns: IndexLabel) -> None:
+    def __init__(
+        self,
+        obj: DataFrame,
+        n: int,
+        keep: Literal["first", "last", "all"],
+        columns: IndexLabel,
+    ) -> None:
         super().__init__(obj, n, keep)
         if not is_list_like(columns) or isinstance(columns, tuple):
             columns = [columns]
@@ -277,4 +298,4 @@ def get_indexer(current_indexer: Index, other_indexer: Index) -> Index:
 
         ascending = method == "nsmallest"
 
-        return frame.sort_values(columns, ascending=ascending, kind="mergesort")
+        return frame.sort_values(columns, ascending=ascending, kind="stable")

pandas/tests/frame/methods/test_nlargest.py

@@ -153,11 +153,11 @@ def test_nlargest_n_duplicate_index(self, n, order, request):
             index=[0, 0, 1, 1, 1],
         )
         result = df.nsmallest(n, order)
-        expected = df.sort_values(order).head(n)
+        expected = df.sort_values(order, kind="stable").head(n)
         tm.assert_frame_equal(result, expected)
 
         result = df.nlargest(n, order)
-        expected = df.sort_values(order, ascending=False).head(n)
+        expected = df.sort_values(order, ascending=False, kind="stable").head(n)
         if Version(np.__version__) >= Version("1.25") and (
             (order == ["a"] and n in (1, 2, 3, 4)) or ((order == ["a", "b"]) and n == 5)
         ):