如何用matplotlib绘制一维高斯混合模型的pdf

时间:2017-02-20 08:37:11

标签: python matplotlib gmm

我想绘制一个高斯混合模型。下面的代码允许我绘制2个单独的高斯,但是在它们相交的地方,线条非常锋利且不够平滑。有没有办法绘制1D GMM的pdf?

def plot_data():
    mu = [-6, 5]
    var = [2, 3]
    sigma = [np.sqrt(var[0]), np.sqrt(var[1])]
    x = np.linspace(-10, 10, 100)
    curve_0 = mlab.normpdf(x, mu[0], sigma[0])
    curve_1 = mlab.normpdf(x, mu[1], sigma[1])
    import ipdb; ipdb.set_trace()
    plt.plot(x, curve_0, color='grey')
    plt.plot(x, curve_1, color='grey')
    plt.fill_between(x,curve_0 , color='grey')
    plt.fill_between(x,curve_1, color='grey')
    plt.show()
    plt.savefig('data_t0.jpg')

2 个答案:

答案 0 :(得分:2)

你可以从高斯混合模型中抽取样本并绘制经验密度/直方图:

import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
n = 10000 # number of sample to be drawn
mu = [-6, 5]
sigma = [2, 3]
samples = []
for i in range(n): # iteratively draw samples
    Z = np.random.choice([0,1]) # latent variable
    samples.append(np.random.normal(mu[Z], sigma[Z], 1))
sns.distplot(samples, hist=False)
plt.show()
sns.distplot(samples)
plt.show()

enter image description here

答案 1 :(得分:0)

你必须形成密度的凸组合

curve = p*curve_0 + (1-p)*curve_1

其中 p 是样本来自第一高斯的概率。