H-K算法的具体解释可自行百度。
import numpy as np
import matplotlib.pyplot as plt
# 输入数据
w1 = np.array([[0, 0, 1], [0, 1, 1], [1, 0, 1]])
w2 = np.array([[1, 1, 1]])
w2 = -w2
b = np.array([1, 1, 1, 1])
c = 1
flag = 0
X = np.concatenate((w1, w2), axis=0)
XX = np.matmul(np.linalg.inv(np.matmul(X.transpose(), X)), X.transpose())
w = np.matmul(XX, b)
e = np.matmul(X, w)-b
t = 0
iteration = 20
while 1:
temp = min(e)
temp1 = max(e)
if 0 > temp > -1e-4:
temp = 0
if temp > 1e-3:
deltab = e + abs(e)
b = b + c*deltab
w = w + c*np.matmul(XX, deltab)
e = np.matmul(X, w) - b
else:
if 1e-4 > temp >= 0:
break
else:
# 线性不可分
if temp1 < 0:
flag = 1
break
else:
# 趋近时迭代
deltab = e + abs(e)
b = b + c * deltab
w = w + c * np.matmul(XX, deltab)
e = np.matmul(X, w) - b
t = t+1
if t >= iteration:
break
print(XX, '\n', w, '\n', e)
x1 = np.arange(0.0, 1.1, .01)
x2 = np.arange(0.0, 1.1, .01)
x1, x2 = np.meshgrid(x1, x2)
f = w[0]*x1+w[1]*x2+w[2]
plt.figure()
plt.contour(x1, x2, f, 0)
plt.scatter(w1[:, 0], w1[:, 1], color="blue")
plt.scatter(-w2[:, 0], -w2[:, 1], color="red")
plt.show()
实验结果: