def kernel_predict(x_out, x_in, y_in, lagrange, kernel):
eps = 1e-5
b = 0
for i in range(len(x_in)):
if lagrange[i] > eps:
b += y_in[i]
for j in range(len(x_in)):
if lagrange[j] > eps:
b -= lagrange[j] * y_in[j] * kernel(x_in[i], x_in[j])
break
y_out = []
for x_out_cur in x_out:
y_cur = 0
for i in range(len(x_in)):
if lagrange[i] > eps:
y_cur += lagrange[i] * y_in[i] * kernel(x_in[i], x_out_cur)
y_cur += b
y_out.append(1.0 if y_cur > 0 else -1.0)
return y_out
def kernel_predict(x_out, x_in, y_in, lagrange, kernel):
eps = 1e-5
b = 0
for i in range(len(x_in)):
if lagrange[i] > eps:
b += y_in[i]
for j in range(len(x_in)):
if lagrange[j] > eps:
b -= lagrange[j] * y_in[j] * kernel(x_in[i], x_in[j])
break
y_out = []
for x_out_cur in x_out:
y_cur = 0
for i in range(len(x_in)):
if lagrange[i] > eps:
y_cur += lagrange[i] * y_in[i] * kernel(x_in[i], x_out_cur)
y_cur += b
y_out.append(1.0 if y_cur > 0 else -1.0)
return y_out
def get_lagrange_coef_reg(x, y, const, kernel):
solvers.options['show_progress'] = False
P = numpy.identity(len(x))
for i in range(len(x)):
for j in range(len(x)):
#P[i][j] = y[i] * y[j] * numpy.inner(x[i], x[j])
P[i][j] = y[i] * y[j] * kernel(x[i], x[j])
P = matrix(P)
A = matrix(transpose([y]))
b = matrix([0.0])
q = matrix([-1.0] * len(x))
G = numpy.zeros(shape=(len(x) * 2, len(x)))
h = matrix([0.0] * len(x) + [const] * len(x))
for i in range(len(x)):
G[i][i] = -1
for i in range(len(x)):
G[len(x) + i][i] = 1
G = matrix(transpose(G))
sol = solvers.qp(P, q, G, h, A, b)['x']
lagrange = [sol[i] for i in range(len(x))]
return lagrange
def get_lagrange_coef_reg(x, y, const, kernel):
solvers.options['show_progress'] = False
P = numpy.identity(len(x))
for i in range(len(x)):
for j in range(len(x)):
#P[i][j] = y[i] * y[j] * numpy.inner(x[i], x[j])
P[i][j] = y[i] * y[j] * kernel(x[i], x[j])
P = matrix(P)
A = matrix(transpose([y]))
b = matrix([0.0])
q = matrix([-1.0] * len(x))
G = numpy.zeros(shape=(len(x) * 2, len(x)))
h = matrix([0.0] * len(x) + [const] * len(x))
for i in range(len(x)):
G[i][i] = -1
for i in range(len(x)):
G[len(x) + i][i] = 1
G = matrix(transpose(G))
sol = solvers.qp(P, q, G, h, A, b)['x']
lagrange = [sol[i] for i in range(len(x))]
return lagrange
