mirror of
https://github.com/TheAlgorithms/Python.git
synced 2024-11-27 23:11:09 +00:00
bc8df6de31
* [pre-commit.ci] pre-commit autoupdate updates: - [github.com/astral-sh/ruff-pre-commit: v0.2.2 → v0.3.2](https://github.com/astral-sh/ruff-pre-commit/compare/v0.2.2...v0.3.2) - [github.com/pre-commit/mirrors-mypy: v1.8.0 → v1.9.0](https://github.com/pre-commit/mirrors-mypy/compare/v1.8.0...v1.9.0) * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
68 lines
1.8 KiB
Python
68 lines
1.8 KiB
Python
"""
|
|
An implementation of interquartile range (IQR) which is a measure of statistical
|
|
dispersion, which is the spread of the data.
|
|
|
|
The function takes the list of numeric values as input and returns the IQR.
|
|
|
|
Script inspired by this Wikipedia article:
|
|
https://en.wikipedia.org/wiki/Interquartile_range
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
|
|
def find_median(nums: list[int | float]) -> float:
|
|
"""
|
|
This is the implementation of the median.
|
|
:param nums: The list of numeric nums
|
|
:return: Median of the list
|
|
>>> find_median(nums=([1, 2, 2, 3, 4]))
|
|
2
|
|
>>> find_median(nums=([1, 2, 2, 3, 4, 4]))
|
|
2.5
|
|
>>> find_median(nums=([-1, 2, 0, 3, 4, -4]))
|
|
1.5
|
|
>>> find_median(nums=([1.1, 2.2, 2, 3.3, 4.4, 4]))
|
|
2.65
|
|
"""
|
|
div, mod = divmod(len(nums), 2)
|
|
if mod:
|
|
return nums[div]
|
|
return (nums[div] + nums[(div) - 1]) / 2
|
|
|
|
|
|
def interquartile_range(nums: list[int | float]) -> float:
|
|
"""
|
|
Return the interquartile range for a list of numeric values.
|
|
:param nums: The list of numeric values.
|
|
:return: interquartile range
|
|
|
|
>>> interquartile_range(nums=[4, 1, 2, 3, 2])
|
|
2.0
|
|
>>> interquartile_range(nums = [-2, -7, -10, 9, 8, 4, -67, 45])
|
|
17.0
|
|
>>> interquartile_range(nums = [-2.1, -7.1, -10.1, 9.1, 8.1, 4.1, -67.1, 45.1])
|
|
17.2
|
|
>>> interquartile_range(nums = [0, 0, 0, 0, 0])
|
|
0.0
|
|
>>> interquartile_range(nums=[])
|
|
Traceback (most recent call last):
|
|
...
|
|
ValueError: The list is empty. Provide a non-empty list.
|
|
"""
|
|
if not nums:
|
|
raise ValueError("The list is empty. Provide a non-empty list.")
|
|
nums.sort()
|
|
length = len(nums)
|
|
div, mod = divmod(length, 2)
|
|
q1 = find_median(nums[:div])
|
|
half_length = sum((div, mod))
|
|
q3 = find_median(nums[half_length:length])
|
|
return q3 - q1
|
|
|
|
|
|
if __name__ == "__main__":
|
|
import doctest
|
|
|
|
doctest.testmod()
|