Merge pull request #20 from MothNik/fix/median_performance

Fix/median performance

Author: deepak7376

.github/workflows/python-package.yml

@@ -27,6 +27,7 @@ jobs:
       run: |
         python -m pip install --upgrade pip
         if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
+        if [ -f requirements-dev.txt ]; then pip install -r requirements-dev.txt; fi
     - name: Run tests
       run: |
         export PYTHONPATH="${PYTHONPATH}:/robustbase/"

.gitignore

@@ -103,12 +103,13 @@ celerybeat.pid
 
 # Environments
 .env
-.venv
+.venv*
 env/
 venv/
 ENV/
 env.bak/
 venv.bak/
+.vscode/
 
 # Spyder project settings
 .spyderproject

README.md

@@ -21,7 +21,7 @@ This package provides functions to calculate the following robust statistical es
 
 ```python
 from robustbase.stats import Qn
-  
+
 x = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
 
 # With bias correction
@@ -37,11 +37,11 @@ res = Qn(x, finite_corr=False)  # result: 4.43828
 
 ```python
 from robustbase.stats import Sn
-  
+
 x = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
 
 # With bias correction
-res = Sn(x)  # result: 3.5778 
+res = Sn(x)  # result: 3.5778
 
 # Without bias correction
 res = Sn(x, finite_corr=False)  # result: 3.5778
@@ -75,7 +75,7 @@ For local development setup:
 ```sh
 git clone https://github.com/deepak7376/robustbase
 cd robustbase
-pip install -r requirements.txt
+pip install -r requirements.txt -r requirements-dev.txt
 ```
 
 ## Recent Changes

requirements-dev.txt

@@ -0,0 +1 @@
+pytest>=8.1.1

requirements.txt

@@ -2,4 +2,3 @@ certifi>=2019.11.28
 docutils>=0.15.2
 numpy>=1.18.0
 statistics>=1.0.3.5
-pytest>=8.1.1

robustbase/__init__.py

@@ -1,6 +1,6 @@
-from .robustbase import Qn
-from .robustbase import Sn
-from .robustbase import iqr
-from .robustbase import mad
+from .robustbase import Qn # noqa: F401
+from .robustbase import Sn # noqa: F401
+from .robustbase import iqr # noqa: F401
+from .robustbase import mad # noqa: F401
 
 

robustbase/stats/__init__.py

@@ -1,4 +1,4 @@
-from .Qn import Qn
-from .Sn import Sn
-from .iqr import iqr
-from .mad import mad 
+from .iqr import iqr  # noqa: F401
+from .mad import mad  # noqa: F401
+from .Qn import Qn  # noqa: F401
+from .Sn import Sn  # noqa: F401

robustbase/utils/__init__.py

@@ -1,2 +1,2 @@
-from .mean import mean
-from .median import median
+from .mean import mean  # noqa: F401
+from .median import median  # noqa: F401

robustbase/utils/median.py

@@ -7,22 +7,29 @@ def median(x, low=False, high=False):
 
     Parameters:
     - x: list or array-like, numeric vector of observations.
-    - low: bool, if True, return the low median for even sample size.
+    - low: bool, if True, return the low median for even sample size. If ``True``, ``high`` is ignored.
     - high: bool, if True, return the high median for even sample size.
 
     Returns:
     - float: Median value.
     """
-    sorted_x = np.sort(x)
-    n = len(sorted_x)
+
+    n = len(x)
     if n == 0:
         raise ValueError("Empty list provided.")
-    
+
+    # for odd sample size, all three medians are the same
     if n % 2 == 1:
-        return sorted_x[n // 2]
-    elif low:
-        return sorted_x[n // 2 - 1]
-    elif high:
-        return sorted_x[n // 2]
-    else:
-        return (sorted_x[n // 2 - 1] + sorted_x[n // 2]) / 2
+        return np.median(a=x)
+
+    # for even sample sizes, the median is the average of the two middle values if
+    # neither the low nor high median is requested
+    if not (low or high):
+        return np.median(a=x)
+
+    # otherwise, either the low or the high median are found via introselect
+    median_idx = n // 2
+    if low:
+        median_idx -= 1
+
+    return np.partition(a=x, kth=median_idx)[median_idx]

tests/test_median.py

@@ -0,0 +1,47 @@
+from typing import Optional, Tuple, Union
+
+import numpy as np
+import pytest
+
+from robustbase.utils.median import median
+
+X_EMPTY = []
+X_ODD_N = [5.5, 3.2, -10.0, -2.1, 8.4]
+X_EVEN_N = [5.5, 3.2, -10.0, -2.1, 8.4, 0.0]
+
+
+@pytest.mark.parametrize("as_array", [False, True])
+@pytest.mark.parametrize(
+    "comb",
+    [
+        (X_EMPTY, False, False, None),
+        (X_EMPTY, True, False, None),
+        (X_EMPTY, False, True, None),
+        (X_EMPTY, True, True, None),
+        (X_ODD_N, False, False, 3.2),
+        (X_ODD_N, True, False, 3.2),
+        (X_ODD_N, False, True, 3.2),
+        (X_ODD_N, True, True, 3.2),
+        (X_EVEN_N, False, False, 1.6),
+        (X_EVEN_N, True, False, 0.0),
+        (X_EVEN_N, False, True, 3.2),
+        (X_EVEN_N, True, True, 0.0),
+    ],
+)
+def test_median(
+    comb: Tuple[Union[list, np.ndarray], bool, bool, Optional[float]],
+    as_array: bool,
+):
+    x, low, high, expected = comb
+    if as_array:
+        x = np.array(x)
+
+    # for empty samples, an error should be raised
+    if expected is None:
+        with pytest.raises(ValueError):
+            median(x, low=low, high=high)
+
+        return
+
+    # otherwise, the expected median should be returned
+    assert median(x, low=low, high=high) == expected