import numpy as np
def createdata():
x1 = np.linspace(0,9,50)
x2 = np.linspace(10,19,50)
x3 = np.linspace(20,9,50)
x = np.concatenate(([x1],[x2],[x3])).T
y = np.dot(x,np.array([1,2,3]).T)
return x,y
def BGD(x,y,learningrate,Max_iter):
x_row_num,dim = x.shape
loss = 1
iter_count = 0
w = np.random.rand(dim)
while loss>0.001 and iter_count<Max_iter:
loss = 0
error = np.zeros(dim)
for i in range(x_row_num):
predict_y = np.dot(w,x[i])
for j in range(dim):
error[j] += (y[i]-predict_y)*x[i][j]#把每一行的误差都累加起来
#求完误差之后更新权重
for i in range(dim):
w[i] += learningrate*error[i]/x_row_num#每一行的都加起来
for i in range(x_row_num):
predict_y = np.dot(w,x[i])
loss += np.square(predict_y-y[i])/(x_row_num*dim)
print('迭代次数:',iter_count,'损失:',loss)
iter_count += 1
return w
if __name__ == '__main__':
x, y = createdata()#构造训练数据
w = BGD(x, y,0.001,10000)#把构造的数据进行梯度下降
print(w) # 会很接近[1, 2,3]