如何基于相关系数生成符合变量间相关性的角度随机值？

Got it, let's solve this problem. Since your three angle variables have a defined correlation structure (as shown in your correlation matrix), picking values independently from each list will break those relationships. Here are two practical approaches to generate random values that preserve the existing correlations:

1. Resample from Existing Correlated Observations #

This is the simplest method if you want to keep the exact distribution and correlation structure of your original data. Instead of sampling from each list separately, you sample entire matched sets of angles from your combined data:

import pandas as pd
import random

# Your original angle data
angles1 = [1,1,5,6,4]
angles2 = [3,2,5,4,9]
angles3 = [6,9,8,2,1]

# Combine into a list of correlated angle tuples
correlated_angle_sets = list(zip(angles1, angles2, angles3))

# Pick a random complete set of correlated angles
random_set = random.choice(correlated_angle_sets)
rand_angle1, rand_angle2, rand_angle3 = random_set

print(f"Correlated random angles: angle1={rand_angle1}, angle2={rand_angle2}, angle3={rand_angle3}")

Pros: #

• Perfectly preserves the original correlation and value distribution
• No complex math required
• Works well if you have enough unique observation sets to sample from

2. Generate New Data with Matching Correlation Structure #

If you need to generate new values beyond your original dataset (not just resample existing ones), you can use a multivariate normal distribution paired with Cholesky decomposition to enforce the correlation structure from your data:

import pandas as pd
import numpy as np

# Original data
angles1 = [1,1,5,6,4]
angles2 = [3,2,5,4,9]
angles3 = [6,9,8,2,1]

# Create DataFrame for easier stats calculation
angles_df = pd.DataFrame({
    'angle1': angles1,
    'angle2': angles2,
    'angle3': angles3
})

# Get mean values and covariance matrix from your data
mean_values = angles_df.mean().values
covariance_matrix = angles_df.cov().values

# Perform Cholesky decomposition to transform normal samples
L = np.linalg.cholesky(covariance_matrix)

# Generate standard normal random values (adjust num_samples for how many sets you need)
num_samples = 1
normal_samples = np.random.normal(size=(num_samples, 3))

# Transform to get samples with your data's mean and covariance
correlated_samples = mean_values + np.dot(normal_samples, L.T)

# Extract the random angles (round to integers if needed, matching your original data)
rand_angle1, rand_angle2, rand_angle3 = np.round(correlated_samples[0]).astype(int)

print(f"Generated correlated angles: angle1={rand_angle1}, angle2={rand_angle2}, angle3={rand_angle3}")

Notes: #

• This method generates values that follow the correlation structure but may create values outside your original data range
• If your angles are circular data (e.g., 0-360 degrees), you’ll need to use circular statistics instead of this normal-based approach (but your sample data looks like non-circular integers, so this should work)
• Rounding to integers (as shown) matches the integer type of your original data

内容的提问来源于stack exchange，提问作者michael