def addpoints(xk, fxk, D, Y2, krig, index, theta, delta, pmax, opt_theta):
points = krig.points
# index(find(index=-1,1)) = length(fvals) + 1
# hard code WARNING
point = Point(2)
point.x = xk.x
point.fval = fxk
points.append(point)
Y2_0 = Y2
index_0 = index
if opt_theta:
raise AssertionError("Optimization of hyperparameters not implemented.")
else:
ntest = 0
psi = 1.0
thetagauss = krig.params[0]
L = []
Y2 = Y2_0
index = index_0
return krig
def rbfmodel(objective, fid, xk, fxk, krig, theta_1, theta_2, theta_3, theta_4, \
delta, deltamax, pmax, optTheta):
deltamax = delta
D = krig.points
Y2, linear, Z, index = affpoints(xk, D, theta_1, theta_3*delta)
if not linear:
Z = Z*delta
for i in range(len(Z[0,:])):
# Hard code alert
point = Point(2)
point.x = xk.x + Z.view(np.ndarray)[:,i]
Y2.append(point)
D.append(point)
f_high = objective(point, fid[1])
point.fval = f_high - objective(point, fid[0])
krig.points.append(point)
index.append(len(krig.points)-1)
linear = True
krig = addpoints(xk, fxk[1]-fxk[0], D, Y2, krig, index, theta_2, theta_4*deltamax, pmax, optTheta)
return krig, linear
def init_models(self):
if self.LHS == 0:
point = Point(self.n)
point.x = np.array([-2.0, 2.0])
f_low = self.objective(point, self.fid[0])
f_high = self.objective(point, self.fid[1])
point.fval = f_high - f_low
self.low_fid_eval += 1
elif self.LHS == 1:
raise AssertionError("Latin hypercube not implemented. Set LHS = 0")
else:
raise AssertionError("LHS can be either 1 or 0")
krig = Kriging(self.corr, self.params, [point])
print point.fval
xk = point
fk = f_high
f_low = f_low
mk = f_high
delta = self.delta_0
krig, linear = rbfmodel(self.objective, self.fid, xk, [f_low, fk], krig,\
self.theta_1, self.theta_2, self.theta_3, self.theta_4,\
delta, self.delta_max, self.pmax, self.optTheta)
