我目前正在學習梯度下降,所以我寫了一段使用線性迴歸梯度下降的代碼。然而,我得到的路線並不是最好的路線。我用梯度下降和最小平方誤差迴歸計算了線性迴歸的誤差。無論使用什麼數據,最小平方誤差總會給我一個更低的誤差。我決定看看這兩個斜坡,並且攔截了兩個。使用梯度下降的y截距總是非常接近於零,就好像它沒有正確改變。我覺得這很奇怪,我不知道發生了什麼。我以某種方式不正確地實現漸變下降?Y截距不會改變線性迴歸梯度下降
import matplotlib.pyplot as plt
datax=[]
datay=[]
def gradient(b_current,m_current,learningRate):
bgradient=0
mgradient=0
N=float(len(datax))
for i in range(0,len(datax)):
bgradient+= (-2/N)*(datay[i]-((m_current*datax[i])+b_current))
mgradient+= (-2/N)*datax[i]*(datay[i]-((m_current*datax[i])+b_current))
newb=b_current-(bgradient*learningRate)
newm=m_current-(mgradient*learningRate)
return newm,newb
def basic_linear_regression(x, y):
# Basic computations to save a little time.
length = len(x)
sum_x = sum(x)
sum_y = sum(y)
# sigma x^2, and sigma xy respectively.
sum_x_squared = sum(map(lambda a: a * a, x))
sum_of_products = sum([x[i] * y[i] for i in range(length)])
# Magic formulae!
a = (sum_of_products - (sum_x * sum_y)/length)/(sum_x_squared - ((sum_x ** 2)/length))
b = (sum_y - a * sum_x)/length
return a, b
def error(m,b,datax,datay):
error=0
for i in range(0,len(datax)):
error+=(datay[i]-(m*datax[i]+b))
return error/len(datax)
def run():
m=0
b=0
iterations=1000
learningRate=.00001
for i in range(0,iterations):
m,b=gradient(b,m,learningRate)
print(m,b)
c,d=basic_linear_regression(datax,datay)
print(c,d)
gradientdescent=error(m,b,datax,datay)
leastsquarederrors=error(c,d,datax,datay)
print(gradientdescent)
print(leastsquarederrors)
plt.scatter(datax,datay)
plt.plot([0,300],[b,300*m+b])
plt.axis('equal')
plt.show()
run()