mirror of
https://github.com/TheAlgorithms/Python.git
synced 2024-11-27 15:01:08 +00:00
b5474ab68a
* Create joint_probability_distribution.py * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update joint_probability_distribution.py * Update joint_probability_distribution.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update maclaurin_series.py * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Revert changes to maclaurin_series.py * Revert changes to maclaurin_series.py * Update joint_probability_distribution.py * Update joint_probability_distribution.py * Update joint_probability_distribution.py --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Christian Clauss <cclauss@me.com>
125 lines
3.9 KiB
Python
125 lines
3.9 KiB
Python
"""
|
|
Calculate joint probability distribution
|
|
https://en.wikipedia.org/wiki/Joint_probability_distribution
|
|
"""
|
|
|
|
|
|
def joint_probability_distribution(
|
|
x_values: list[int],
|
|
y_values: list[int],
|
|
x_probabilities: list[float],
|
|
y_probabilities: list[float],
|
|
) -> dict:
|
|
"""
|
|
>>> joint_distribution = joint_probability_distribution(
|
|
... [1, 2], [-2, 5, 8], [0.7, 0.3], [0.3, 0.5, 0.2]
|
|
... )
|
|
>>> from math import isclose
|
|
>>> isclose(joint_distribution.pop((1, 8)), 0.14)
|
|
True
|
|
>>> joint_distribution
|
|
{(1, -2): 0.21, (1, 5): 0.35, (2, -2): 0.09, (2, 5): 0.15, (2, 8): 0.06}
|
|
"""
|
|
return {
|
|
(x, y): x_prob * y_prob
|
|
for x, x_prob in zip(x_values, x_probabilities)
|
|
for y, y_prob in zip(y_values, y_probabilities)
|
|
}
|
|
|
|
|
|
# Function to calculate the expectation (mean)
|
|
def expectation(values: list, probabilities: list) -> float:
|
|
"""
|
|
>>> from math import isclose
|
|
>>> isclose(expectation([1, 2], [0.7, 0.3]), 1.3)
|
|
True
|
|
"""
|
|
return sum(x * p for x, p in zip(values, probabilities))
|
|
|
|
|
|
# Function to calculate the variance
|
|
def variance(values: list[int], probabilities: list[float]) -> float:
|
|
"""
|
|
>>> from math import isclose
|
|
>>> isclose(variance([1,2],[0.7,0.3]), 0.21)
|
|
True
|
|
"""
|
|
mean = expectation(values, probabilities)
|
|
return sum((x - mean) ** 2 * p for x, p in zip(values, probabilities))
|
|
|
|
|
|
# Function to calculate the covariance
|
|
def covariance(
|
|
x_values: list[int],
|
|
y_values: list[int],
|
|
x_probabilities: list[float],
|
|
y_probabilities: list[float],
|
|
) -> float:
|
|
"""
|
|
>>> covariance([1, 2], [-2, 5, 8], [0.7, 0.3], [0.3, 0.5, 0.2])
|
|
-2.7755575615628914e-17
|
|
"""
|
|
mean_x = expectation(x_values, x_probabilities)
|
|
mean_y = expectation(y_values, y_probabilities)
|
|
return sum(
|
|
(x - mean_x) * (y - mean_y) * px * py
|
|
for x, px in zip(x_values, x_probabilities)
|
|
for y, py in zip(y_values, y_probabilities)
|
|
)
|
|
|
|
|
|
# Function to calculate the standard deviation
|
|
def standard_deviation(variance: float) -> float:
|
|
"""
|
|
>>> standard_deviation(0.21)
|
|
0.458257569495584
|
|
"""
|
|
return variance**0.5
|
|
|
|
|
|
if __name__ == "__main__":
|
|
from doctest import testmod
|
|
|
|
testmod()
|
|
# Input values for X and Y
|
|
x_vals = input("Enter values of X separated by spaces: ").split()
|
|
y_vals = input("Enter values of Y separated by spaces: ").split()
|
|
|
|
# Convert input values to integers
|
|
x_values = [int(x) for x in x_vals]
|
|
y_values = [int(y) for y in y_vals]
|
|
|
|
# Input probabilities for X and Y
|
|
x_probs = input("Enter probabilities for X separated by spaces: ").split()
|
|
y_probs = input("Enter probabilities for Y separated by spaces: ").split()
|
|
assert len(x_values) == len(x_probs)
|
|
assert len(y_values) == len(y_probs)
|
|
|
|
# Convert input probabilities to floats
|
|
x_probabilities = [float(p) for p in x_probs]
|
|
y_probabilities = [float(p) for p in y_probs]
|
|
|
|
# Calculate the joint probability distribution
|
|
jpd = joint_probability_distribution(
|
|
x_values, y_values, x_probabilities, y_probabilities
|
|
)
|
|
|
|
# Print the joint probability distribution
|
|
print(
|
|
"\n".join(
|
|
f"P(X={x}, Y={y}) = {probability}" for (x, y), probability in jpd.items()
|
|
)
|
|
)
|
|
mean_xy = expectation(
|
|
[x * y for x in x_values for y in y_values],
|
|
[px * py for px in x_probabilities for py in y_probabilities],
|
|
)
|
|
print(f"x mean: {expectation(x_values, x_probabilities) = }")
|
|
print(f"y mean: {expectation(y_values, y_probabilities) = }")
|
|
print(f"xy mean: {mean_xy}")
|
|
print(f"x: {variance(x_values, x_probabilities) = }")
|
|
print(f"y: {variance(y_values, y_probabilities) = }")
|
|
print(f"{covariance(x_values, y_values, x_probabilities, y_probabilities) = }")
|
|
print(f"x: {standard_deviation(variance(x_values, x_probabilities)) = }")
|
|
print(f"y: {standard_deviation(variance(y_values, y_probabilities)) = }")
|