__author__ = 'fengchen'
from cvxopt import matrix, solvers, spmatrix
import numpy as np
import matplotlib.pyplot as plt
from math import *
def calc_Q_p_G_h(data):
# Q = spmatrix(2.0, range(3), range(3))
Q = matrix([0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], (3,3))
Q[0,0] = 2
Q[1,1] = 2
p = matrix([0.0, 0.0, 0.0], (3,1))
G = []
h = []
for items in data:
row = []
if items[2] == 1:
row.extend([-1 * item for item in items[:2]])
row.append(-1)
G.append(row)
h.append(-1.0)
else:
row.extend(items[:2])
row.append(1)
G.append(row)
h.append(-1.0)
#print G
G = matrix(G).trans()
h = matrix(h)
print(Q, p)
print(G, h)
print(type(G))
return Q, p, G, h
def plot(data, w1, w2, b, sol):
### suppose you have obtained sol['x'] from CVXOPT QP minimization
x = [item[0] for item in data if item[2] == 1]
y = [item[1] for item in data if item[2] == 1]
plt.scatter(x, y, s=80, facecolors='none', edgecolors='r')
x = [item[0] for item in data if item[2] == -1]
y = [item[1] for item in data if item[2] == -1]
plt.scatter(x, y, s=80, facecolors='none', edgecolors='b')
x = [item[0] for item in data]
y = [item[1] for item in data]
plt.scatter(x, y, s=40, facecolors='none', edgecolors='k')
w1 = sol['x'][0]
w2 = sol['x'][1]
b = sol['x'][2]
print('w: {0}; b: {1} \n'.format([w1,w2], b))
print('w.x + b > 0 for positive class')
print('w.x + b < 0 for negative class')
x = [-1, 6]
y = [(w1 * x[i] + b)/(-1 * w2) for i in range(2)]
plt.plot(x, y, color='red')
y = [(w1 * x[i] + b - 1)/(-1 * w2) for i in range(2)]
plt.plot(x, y, color='black')
y = [(w1 * x[i] + b + 1)/(-1 * w2) for i in range(2)]
plt.plot(x, y, color='black')
plt.gca().set_aspect('equal', adjustable='box')
plt.grid()
plt.gcf().set_size_inches(5, 5)
plt.show()
def svm_proc(data):
Q, p, G, h = calc_Q_p_G_h(data)
sol=solvers.qp(Q, p, G, h)
w1 = sol['x'][0]
w2 = sol['x'][1]
b = sol['x'][2]
print('\n\nw1= {0}; w2={1}; b={2}'.format(w1, w2, b))
plot(data, w1, w2, b, sol)
