数据分析——两种求解R平方的方法

1、皮尔逊相关系数 (Pearson Correlation Coefficient)是衡量两个值线性相关强度的量，取值范围：[-1, 1]，正向相关：>0，负向相关：<0，无相关性：=0

上式又可以表示为：

R^2是皮尔逊相关系数的平方，依然是表示两个值线性相关强度的量，取值范围：[0, 1]，值越大，相关性越强。

import numpy as np
 
def computeCorrelation(x, y):
	xBar = np.mean(x)
	yBar = np.mean(y)
	covXY = 0
	varX = 0
	varY = 0
	for i in range(0, len(x)):
		diffxx = x[i] - xBar
		diffyy = y[i] - yBar
		covXY += (diffxx * diffyy)
		varX += diffxx ** 2
		varY += diffyy ** 2
	return covXY / np.sqrt(varX * varY)

x=np.linspace(-0.5,0.5,10)
noise=np.random.normal(0,0.02,10)
y=2*x+1+noise

R = computeCorrelation(x, y)
R2 = np.square(R)
print(R2)

运行结果：
0.9994448592313179

2、R^2还可以由下式计算得到，yihat表示预测值，yi表示真实值，ybar表示真实值的均值

import numpy as np

def computeCorrelation(x, y, degree):
	result = {}
	coeffs = np.polyfit(x, y, degree) # degree表示N次拟合
	result['coeffs'] = coeffs.tolist()
 
	p = np.poly1d(coeffs) # 得到拟合方程
	yhat = p(x) # 得到预测值
	ybar = np.sum(y) / len (y)
	SSR = np.sum((yhat -ybar)**2)
	SST = np.sum((y - ybar)**2)
	result['R2'] = SSR / SST
 
	return result['R2']

x=np.linspace(-0.5,0.5,10)
noise=np.random.normal(0,0.02,10)
y=2*x+1+noise

R2 = computeCorrelation(x, y, 1)
print(R2)

运行结果：
0.999732826660141