Python/searches/median_of_medians.py

108 lines
2.7 KiB
Python
Raw Normal View History

"""
A Python implementation of the Median of Medians algorithm
to select pivots for quick_select, which is efficient for
calculating the value that would appear in the index of a
list if it would be sorted, even if it is not already
sorted. Search in time complexity O(n) at any rank
deterministically
https://en.wikipedia.org/wiki/Median_of_medians
"""
def median_of_five(arr: list) -> int:
"""
Return the median of the input list
:param arr: Array to find median of
:return: median of arr
>>> median_of_five([2, 4, 5, 7, 899])
5
>>> median_of_five([5, 7, 899, 54, 32])
32
>>> median_of_five([5, 4, 3, 2])
4
>>> median_of_five([3, 5, 7, 10, 2])
5
"""
arr = sorted(arr)
return arr[len(arr) // 2]
def median_of_medians(arr: list) -> int:
"""
Return a pivot to partition data on by calculating
Median of medians of input data
:param arr: The data to be checked (a list)
:return: median of medians of input array
>>> median_of_medians([2, 4, 5, 7, 899, 54, 32])
54
>>> median_of_medians([5, 7, 899, 54, 32])
32
>>> median_of_medians([5, 4, 3, 2])
4
>>> median_of_medians([3, 5, 7, 10, 2, 12])
12
"""
if len(arr) <= 5:
return median_of_five(arr)
medians = []
i = 0
while i < len(arr):
if (i + 4) <= len(arr):
medians.append(median_of_five(arr[i:].copy()))
else:
medians.append(median_of_five(arr[i : i + 5].copy()))
i += 5
return median_of_medians(medians)
def quick_select(arr: list, target: int) -> int:
"""
Two way partition the data into smaller and greater lists,
in relationship to the pivot
:param arr: The data to be searched (a list)
:param target: The rank to be searched
:return: element at rank target
>>> quick_select([2, 4, 5, 7, 899, 54, 32], 5)
32
>>> quick_select([2, 4, 5, 7, 899, 54, 32], 1)
2
>>> quick_select([5, 4, 3, 2], 2)
3
>>> quick_select([3, 5, 7, 10, 2, 12], 3)
5
"""
# Invalid Input
if target > len(arr):
return -1
# x is the estimated pivot by median of medians algorithm
x = median_of_medians(arr)
left = []
right = []
check = False
for i in range(len(arr)):
if arr[i] < x:
left.append(arr[i])
elif arr[i] > x:
right.append(arr[i])
elif arr[i] == x and not check:
check = True
else:
right.append(arr[i])
rank_x = len(left) + 1
if rank_x == target:
answer = x
elif rank_x > target:
answer = quick_select(left, target)
elif rank_x < target:
answer = quick_select(right, target - rank_x)
return answer
print(median_of_five([5, 4, 3, 2]))