使用Matplotlib绘制Dirichlet分布热力图出现异常图像的原因

Hey there! Let's break down why your heatmap isn't turning out as expected, and fix it up.

First, let's look at what your code is actually doing:

• np.random.dirichlet(alpha=[0.3, 0.7], size=1000) generates a (1000, 2) array. Every row in this array sums to 1 (that's the defining property of the Dirichlet distribution on a 2-simplex—think of it as points lying on the line x1 + x2 = 1 where x1, x2 ≥ 0).
• plt.imshow() is designed to visualize 2D matrices (images). When you pass a (1000,2) array to it, it treats each row as a horizontal line of 2 pixels, resulting in a weird "two vertical strips" plot. This has nothing to do with the actual shape of the Dirichlet distribution—you're using the wrong tool for the job here.

Fix 1: Visualize 2D Dirichlet with a Histogram #

Since a 2D Dirichlet reduces to a 1D distribution (because x2 = 1 - x1), the simplest way to show its shape is with a histogram of one of the variables:

import numpy as np
import matplotlib.pyplot as plt

a = np.random.dirichlet(alpha=[0.3, 0.7], size=1000)
# Grab the first variable (the second would just be its mirror)
x = a[:, 0]

plt.hist(x, bins=30, density=True, alpha=0.6, color='#e74c3c')
plt.xlabel('$x_1$')
plt.ylabel('Density')
plt.title('Dirichlet Distribution (α = [0.3, 0.7])')
plt.show()

Fix 2: Plot Samples on the 2-Simplex #

If you want to see where the samples lie geometrically, you can plot them on the line x1 + x2 = 1:

import numpy as np
import matplotlib.pyplot as plt

a = np.random.dirichlet(alpha=[0.3, 0.7], size=1000)
x1 = a[:, 0]
x2 = a[:, 1]

plt.scatter(x1, x2, s=6, alpha=0.5, color='#3498db')
# Draw the simplex boundary
plt.plot([0, 1], [1, 0], 'k--', linewidth=1)
plt.xlim(0, 1)
plt.ylim(0, 1)
plt.xlabel('$x_1$')
plt.ylabel('$x_2$')
plt.title('Dirichlet Samples on the 2-Simplex')
plt.show()

If You Really Want a Heatmap (Use 3D Dirichlet) #

Heatmaps make sense for higher-dimensional Dirichlet distributions (3D or more), where you can plot the joint distribution of the first two variables (since the third is 1 - x1 - x2). Here's how to do that:

import numpy as np
import matplotlib.pyplot as plt

# Generate samples from a 3D Dirichlet distribution
a = np.random.dirichlet(alpha=[0.3, 0.7, 1.0], size=10000)
x1 = a[:, 0]
x2 = a[:, 1]

# Create a 2D histogram heatmap
plt.hist2d(x1, x2, bins=35, cmap='hot')
plt.colorbar(label='Sample Count')
plt.xlabel('$x_1$')
plt.ylabel('$x_2$')
plt.title('Heatmap of 3D Dirichlet Distribution')
plt.show()

To recap:

• A 2D Dirichlet is a 1D distribution, so imshow (a 2D image tool) is not the right choice here.
• Use histograms or simplex scatter plots for 2D Dirichlet.
• Reserve heatmaps for 3D+ Dirichlet distributions, where you can visualize the joint density of two variables.

内容的提问来源于stack exchange，提问作者user8270077