分布函数求导就是概率密度,在任意的数据中由于数据分布不确定,所以严格来讲这些数据是没有分布函数和概率密度函数的,但是我们就是想得出那种结果,画出那种曲线,这里就需要借助scipy.stats.relfreq
这个方法,它可以拟合这样的结果
得到CDF、PDF曲线对应的值
需要导包:
import numpy as npimport matplotlib.pyplot as pltfrom scipy import stats%matplotlib inline
首先构造数据:
data = np.random.normal(0,10,100) # 生成100个随机数,这里生成正态分布,任意分布都行,正态分布效果更明显一些res_freq = stats.relfreq(data, numbins=20) # numbins 是统计一次的间隔(步长)是多大
概率密度PDF
pdf_value = res_freq.frequency
累积分布CDF
cdf_value = np.cumsum(res_freq.frequency)
开始绘图
首先确定横坐标,这里横坐标需要使用线性计算计算得出,不能直接调属性:
x = res_freq.lowerlimit + np.linspace(0, res_freq.binsize * res_freq.frequency.size, res_freq.frequency.size)
PDF的图像
plt.bar(x, pdf_value, width=res_freq.binsize)
CDF的图像
plt.plot(x, cdf_value)
完整代码
import numpy as npimport matplotlib.pyplot as pltfrom scipy import stats%matplotlib inline# 构造数据data = np.random.normal(0,10,100)res_freq = stats.relfreq(data, numbins=100)# 计算结果pdf_value = res_freq.frequencycdf_value = np.cumsum(res_freq.frequency)# 绘图x = res_freq.lowerlimit + np.linspace(0, res_freq.binsize * res_freq.frequency.size, res_freq.frequency.size)plt.bar(x, pdf_value, width=res_freq.binsize)plt.plot(x, cdf_value)