def Mutation2(Members, NM, Scale):
GenesForMutation = np.random.randint(0, len(Members), NM)
MutatedMembers = np.copy(Members)
for i in GenesForMutation:
gene = np.random.randint(0, len(Members[0]), 1)
if Members[i, gene] >= 0:
MutationP = np.random.uniform(Members[i, gene],
Members[i, gene] * (1 + Scale), 1)
MutationM = np.random.uniform(Members[i, gene] * (1 - Scale),
Members[i, gene], 1)
else:
MutationP = np.random.uniform(Members[i, gene],
Members[i, gene] * (1 - Scale), 1)
MutationM = np.random.uniform(Members[i, gene] * (1 + Scale),
Members[i, gene], 1)
MutatedP = MutatedMembers[i]
MutatedP[gene] = MutationP
frosenP = rosen(MutatedP)
#print(frosenP)
MutatedM = MutatedMembers[i]
MutatedM[gene] = MutationM
frosenM = rosen(MutatedM)
#print(frosenM)
if frosenP >= frosenM:
MutatedMembers[i, gene] = MutationM
#print(rosen(MutatedMembers[i]))
if frosenM > frosenP:
MutatedMembers[i, gene] = MutationP
#print(rosen(MutatedMembers[i]))
return MutatedMembers
def plot(self, function, type = 'surface'):
"""
Plots a given function. Type defines what kind of class of plots to plot,
the alternatives are; surface plot or a contour plot, it defaults to a
surface plot.
"""
if (type == 'surface'):
x = linspace(-2,2,250) #Defines plot intervall for x direction.
y = linspace(-1,3,250) #Defines plot intervall for y direction.
X, Y = meshgrid(x, y)
Z = rosen([X, Y]) #Defines plot intervall for z direction.
ax = plt.axes(projection='3d')
ax.plot_surface(X,Y,Z, norm = LogNorm(), rstride = 5, cstride = 5, cmap = 'RdGy_r',
alpha = .9, edgecolor = 'none')
ax.set_title('Rosenbrock function - Surface plot')
ax.set_xlabel('x_1')
ax.set_ylabel('x_2')
ax.set_zlabel('f(x)')
if (type == 'contour'):
ax = plt.axes()
x = linspace(-5,5,500) #Defines plot intervall for x direction.
y = linspace(-5,5,500) #Defines plot intervall for y direction.
X, Y = meshgrid(x, y)
Z = rosen([X, Y])
plt.contour(X, Y, Z, 150, cmap = 'RdGy');
ax.set_title('Rosenbrock function - Contour plot')
ax.set_xlabel('x_1')
ax.set_ylabel('x_2')
plt.show()
def val(self, x, test=False):
assert len(x) == self.N
val = 0
for idx in range(len(x) - 1):
val += (1 - x[idx])**2 + 100 * (x[idx + 1] - (x[idx])**2)**2
if test:
if val != rosen(x):
np.testing.assert_almost_equal(val, rosen(x))
return val
def rosen_obj(params, shift):
val = rosen(params["x_half_a"] -
shift) + rosen(params["x_half_b"] - shift)
dval = OrderedDict([
("x_half_a", rosen_der(params["x_half_a"] - shift)),
("x_half_b", rosen_der(params["x_half_b"] - shift)),
])
return val, dval
def generate_rosen_data(size, d, meshgrid=False):
if not meshgrid:
X = np.random.rand(size, d)
y = rosen(X.T)[:, None]
return X, y
else:
Xi, Xj = np.meshgrid(np.linspace(0, 1), np.linspace(0, 1))
X = np.vstack([Xi.ravel(), Xj.ravel()]).T
y = rosen(X.T)
Y = y.reshape(Xi.shape)
return X, y[:, None], Xi, Xj, Y
def plot_trace(ps, ttl):
x = np.linspace(-5, 5, 100)
y = np.linspace(-5, 5, 100)
X, Y = np.meshgrid(x, y)
Z = rosen(np.vstack([X.ravel(), Y.ravel()])).reshape((100,100))
ps = np.array(ps)
plt.figure(figsize=(12,4))
plt.subplot(121)
plt.contour(X, Y, Z, np.arange(10)**5)
plt.plot(ps[:, 0], ps[:, 1], '-o')
plt.plot(1, 1, 'r*', markersize=12) # global minimum
plt.subplot(122)
plt.semilogy(range(len(ps)), rosen(ps.T))
plt.title(ttl)
def plot_trace(ps, ttl):
x = np.linspace(-5, 5, 100)
y = np.linspace(-5, 5, 100)
X, Y = np.meshgrid(x, y)
Z = rosen(np.vstack([X.ravel(), Y.ravel()])).reshape((100, 100))
ps = np.array(ps)
plt.figure(figsize=(12, 4))
plt.subplot(121)
plt.contour(X, Y, Z, np.arange(10)**5)
plt.plot(ps[:, 0], ps[:, 1], '-o')
plt.plot(1, 1, 'r*', markersize=12) # global minimum
plt.subplot(122)
plt.semilogy(list(range(len(ps))), rosen(ps.T))
plt.title(ttl)
def rosenbrock(x):
"""The Rosenbrock funtion is is a non-convex function used as a performance test problem for optimization
algorithms. The function is defined by: f(x,y) = (a-x)^2 + b(y-x^2)^2. The global minimum is inside a long, narrow,
parabolic shaped flat valley. To find the valley is trivial. To converge to the global minimum, is difficult.
More info: https://en.wikipedia.org/wiki/Rosenbrock_function
"""
return rosen(x)
def test_tqdm_resume_interrupted():
des = DESolver(lambda x: [rosen(x), time.sleep(.001)][0],
bounds=[(0, 2), (0, 2), (0, 2), (0, 2), (0, 2)],
popsize=2, maxfun=23, p_bars=True)
des.solve()
assert des.pbar_feval.n == 24
assert des.pbar_gen_mutations.n == 4
assert np.isclose(des.pbar_gens.n, 1.0 + 4 / 10)
state = des.state.copy()
state['maxfun'] = 40
des = DESolver(rosen, **state)
assert des._nfev == 24
assert des.pbar_feval.n == 24
assert des.pbar_gen_mutations.n == 4
assert np.isclose(des.pbar_gens.n, 1.0 + 4 / 10)
x, y = next(des)
assert des._nfev == 34
assert des.pbar_feval.n == 34
assert des.pbar_gen_mutations.n == 4
assert np.isclose(des.pbar_gens.n, 2.0+4/10)
with pytest.raises(StopIteration):
x, y = next(des)
assert des._nfev == 41
assert des.pbar_feval.n == 41
assert des.pbar_gen_mutations.n == 1
assert np.isclose(des.pbar_gens.n, 3.0+1/10)
def testMaximization(self):
self.optimizer.isMaximize = True
rosen_inv = lambda x: - rosen(x)
rosen_der_inv = lambda x: - rosen_der(x)
max_params, max_value, _ = self.optimizer.optimize(rosen_inv, rosen_der_inv, x0=self.x0)
self.assertAlmostEqual(self.reference_value, max_value, 10)
self.assertAlmostEqual(self.reference_params[0], max_params[0], 4)
self.assertAlmostEqual(self.reference_params[1], max_params[1], 4)
def calc_obj_func(sol, obj_func='sphere'):
result = 0
if obj_func == 'sphere':
result = sphere(sol)
elif obj_func == 'rosenbrock':
result = rosen(sol)
else:
return 0
return result
def nltest(x,grad):
nonlocal counter
nonlocal countergrad
if len(grad) > 0:
countergrad += 1
grad[:] = rosen_der(x)
return grad
counter += 1
return rosen(x)
def compute(self, inputs, outputs):
time.sleep(self.options['sleep_time'])
nan_points = self.options['nan_points']
if nan_points is not None:
for nan_point in nan_points:
if np.linalg.norm(inputs['x'] - np.asarray(nan_point)
) <= self.options['nan_range']:
outputs['f'] = 1e27
return
outputs['f'] = rosen(inputs['x'])
def test_tqdm():
des = DESolver(lambda x: [rosen(x), time.sleep(.001)][0],
bounds=[(0, 2), (0, 2), (0, 2), (0, 2), (0, 2)],
popsize=2, maxiter=2, maxfun=30, p_bars=True)
des.solve()
assert des.pbar_feval.n == 30
assert des.pbar_gen_mutations.n == 0
assert np.isclose(des.pbar_gens.n, 2.0)
des = DESolver(lambda x: [rosen(x), time.sleep(.001)][0],
bounds=[(0, 2), (0, 2), (0, 2), (0, 2), (0, 2)],
popsize=2, maxiter=1, maxfun=30, p_bars=True)
des.solve()
assert des.pbar_feval.n == 20
assert des.pbar_gen_mutations.n == 0
assert np.isclose(des.pbar_gens.n, 1.0)
des = DESolver(lambda x: [rosen(x), time.sleep(.001)][0],
bounds=[(0, 2), (0, 2), (0, 2), (0, 2), (0, 2)],
popsize=2, maxiter=2, p_bars=True)
des.solve()
assert des.pbar_feval.n == 30
assert des.pbar_gen_mutations.n == 0
assert np.isclose(des.pbar_gens.n, 2.0)
des = DESolver(lambda x: [rosen(x), time.sleep(.001)][0],
bounds=[(0, 2), (0, 2), (0, 2), (0, 2), (0, 2)],
popsize=2, maxfun=23, p_bars=True)
des.solve()
assert des.pbar_feval.n == 24
assert des.pbar_gen_mutations.n == 4
assert np.isclose(des.pbar_gens.n, 1.0 + 4 / 10)
des = DESolver(lambda x: [rosen(x), time.sleep(.001)][0],
bounds=[(0, 2), (0, 2), (0, 2), (0, 2), (0, 2)],
popsize=2, maxiter=1, p_bars=True)
des.solve()
assert des.pbar_feval.n == 20
assert des.pbar_gen_mutations.n == 0
assert np.isclose(des.pbar_gens.n, 1.0)
def _build_set(self, use_cache, rewrite):
if use_cache:
try:
return self._load()
except AssertionError:
print("Can't use cache, generating new dataset")
x = 2*np.random.random((self.data_size, self.n_dim)) - 1
y = rosen(x.T)[:, None]
if rewrite:
self._save('x', x)
self._save('y', y)
return x, y
def test_minimize_ipopt_nojac_constraints_if_scipy():
""" `minimize_ipopt` works without Jacobian and with constraints"""
from scipy.optimize import rosen, rosen_der
x0 = [1.3, 0.7, 0.8, 1.9, 1.2]
constr = {"fun": lambda x: rosen(x) - 1.0, "type": "ineq"}
res = cyipopt.minimize_ipopt(rosen, x0, jac=rosen_der, constraints=constr)
assert isinstance(res, dict)
assert np.isclose(res.get("fun"), 1.0)
assert res.get("status") == 0
assert res.get("success") is True
expected_res = np.array(
[1.00407015, 0.99655763, 1.05556205, 1.18568342, 1.38386505])
np.testing.assert_allclose(res.get("x"), expected_res)
def evaluate(self, x):
assert len(x) == self.d
x = x.astype(np.int64)
# Compute non-binary representation
unbinarized_x = []
for i in range(self.unbinarized_d):
x_i = 0
for j in x[4 * i:4 * i + 4]:
x_i = (x_i << 1) | j # bitshift to convert binary to integer
normalized_x = x_i + self.unbinarized_lb
unbinarized_x.append(normalized_x)
return rosen(unbinarized_x) / self.scaling
def test_minimize_ipopt_nojac_constraints_if_scipy():
""" `minimize_ipopt` works without Jacobian and with constraints"""
from scipy.optimize import rosen
x0 = [1.3, 0.7, 0.8, 1.9, 1.2]
constr = {"fun": lambda x: rosen(x) - 1.0, "type": "ineq"}
res = cyipopt.minimize_ipopt(rosen, x0, constraints=constr)
assert isinstance(res, dict)
assert np.isclose(res.get("fun"), 1.0)
assert res.get("status") == 0
assert res.get("success") is True
expected_res = np.array([1.001867, 0.99434067, 1.05070075, 1.17906312,
1.38103001])
np.testing.assert_allclose(res.get("x"), expected_res)
def test_optimization_trust():
print("**********************************************")
print("TEST Newton trust region ")
x0 = [1.3, 0.7, 0.8, 1.9, 1.2]
res = optimize.minimize(
optimize.rosen, x0, method='trust-ncg',
jac=optimize.rosen_der,
hess=optimize.rosen_hess,
options={'gtol': 1e-8, 'disp': True})
print(res.x)
print(optimize.rosen(x0).shape)
print(optimize.rosen_der(x0).shape)
print(optimize.rosen_hess(x0).shape)
return res.fun
def rosenbrockLnlike(theta):
"""
Rosenbrock function as a loglikelihood following Wang & Li (2017)
Parameters
----------
theta : array
Returns
-------
l : float
likelihood
"""
return -rosen(theta) / 100.0
def test(self):
# let's at least make sure it doesn't error
el_test = EvaluationLogger("Test logger")
for i in range(50):
arg = [0.0, 0.0]
el_test.push([arg], rosen(arg))
min_coord = [-3, -3]
max_coord = [3, 3]
convergence_demo2d([el_test], rosen, min_coord, max_coord,
resample_evals_to=25, show_plot=False)
self.assertTrue(True)
goal_val_plot([el_test], mintoi=True, xtimestep='time',
fmin=0, plot_logarithm=False, show_plot=False)
self.assertTrue(True)
def rosenbrock(x):
"""The Rosenbrock funtion is is a non-convex function used as a performance test problem for optimization
algorithms. The function is defined by: f(x,y) = (a-x)^2 + b(y-x^2)^2. The global minimum is inside a long, narrow,
parabolic shaped flat valley. To find the valley is trivial. To converge to the global minimum, is difficult.
More info: https://en.wikipedia.org/wiki/Rosenbrock_function
Examples:
>>> bounds = [(0,2), (0, 2), (0, 2), (0, 2), (0, 2)]
>>> result = optimize_function(rosenbrock, bounds)
>>> result.x # Solution
array([ 1., 1., 1., 1., 1.])
>>> result.fun # Final value of the objective function
0.0
>>> result.success
True
"""
return rosen(x)
def dim53Rosenbrock(ht_list, x):
XX = []
assert len(ht_list) == 50
h2 = [
1, 2
] #convert to these categories, as Cocabo assumes categories in (0,1,2,3,etc.)
for i in ht_list:
if i:
XX.append(h2[i])
else:
XX.append(h2[0])
for i in x:
XX.append(i)
XX[0:len(ht_list)] = np.round(
XX[0:len(ht_list)]
) #To make sure there is no cheating, round the discrete variables before calling the function
return rosen(XX) / 20000 + 1e-6 * np.random.rand()
def test_tqdm_resume():
des = DESolver(lambda x: [rosen(x), time.sleep(.001)][0],
bounds=[(0, 2), (0, 2), (0, 2), (0, 2), (0, 2)],
popsize=2, maxiter=1, p_bars=True)
des.solve()
assert des.pbar_feval.n == 20
assert des.pbar_gen_mutations.n == 0
assert np.isclose(des.pbar_gens.n, 1.0)
des = DESolver(rosen, **des.state)
assert des._nfev == 20
assert des.pbar_feval.n == 20
assert des.pbar_gen_mutations.n == 0
assert des.pbar_gens.n == 1.0+0/10
x, y = next(des)
assert des._nfev == 30
assert des.pbar_feval.n == 30
assert des.pbar_gen_mutations.n == 0
assert np.isclose(des.pbar_gens.n, 2.0+0/10)
def f(x):
time.sleep(0.1)
return [opt.rosen(x), opt.rosen_der(x)]
def __call__(self, x):
self.x = x
self.fun = optimize.rosen(x)
self.nit += 1
def func_grad_hess(x,*args):
f = optimize.rosen(x)
g = optimize.rosen_der(x)
h= optimize.rosen_hess(x)
return (f,g,h)
def f(x):
time.sleep(0.01)
return [opt.rosen(x), opt.rosen_der(x)]
def fun(self, x, *args):
self.nfev += 1
return rosen(x)
def fun(self, x):
time.sleep(40e-6)
return rosen(x)
def evaluate(self, x):
assert len(x) == self.d
if self.dolog:
return np.log(rosen(x) + 1) + self.noise_rng.normal(scale=self.noise_factor)
return rosen(x)/self.scaling + self.noise_rng.normal(scale=self.noise_factor)
def myrosen(pars):
return rosen(pars.values())
def SysEq(x, gc):
f = rosen(x)
return f, []
optlog.callback(x0)
try:
xfin = opt.minimize(f, jac=True, x0=x0, method='CG', callback=optlog.callback, options={'gtol': 0.0})
xfin = opt.minimize(f, jac=True, x0=xfin.x, method='CG', callback=optlog.callback)
xfin = opt.minimize(f, jac=True, x0=xfin.x, method='CG', callback=optlog.callback)
finx = xfin.x
optlog.finish(finx)
print(xfin)
except KeyboardInterrupt:
finx = optlog.hist.iloc[-1, :].x
print(finx)
hist = pd.read_pickle('./opthist.pkl')
plt.figure()
plt.plot(hist.i, hist.f, '-x')
plt.plot(hist.i, hist.gnorm, '-x')
plt.xscale('log')
plt.yscale('log')
plt.figure()
X, Y = np.meshgrid(np.linspace(-5, 5, 500),
np.linspace(-5, 5, 500))
xy = np.hstack((X.flatten()[:, None], Y.flatten()[:, None]))
feval = np.array([opt.rosen(v) for v in xy]).reshape(len(X), len(Y))
plt.contour(X, Y, np.log(feval))
hist_points = np.vstack(hist.x)
plt.plot(hist_points[:, 0], hist_points[:, 1], 'x-')
plt.show()
def test_function_selection5():
forrester = lambda x: (5.0*x-2.0)**2.0*np.sin(12.*x-4.)
# def f(x):
# x = np.array(x)*0.08
# return forrester(np.linalg.norm(x)) + 5.*np.linalg.norm(x) + 10.*(x[0]+x[1])
# def f2(x):
# x = np.array(x)*0.085
# return forrester(np.linalg.norm(x))
# def f(x):
#
# return forrester(np.linalg.norm(0.08*np.array(x))) + 0.9*np.linalg.norm(x) + 0.2*(x[0]+x[1])
# def f2(x):
# x[0] = x[0]+1.0
# x[1] = x[1]+1.0
# return forrester(np.linalg.norm(0.035*np.array(x))) + 0.5*np.linalg.norm(x) + 0.1*(x[0]+x[1])
f = lambda x: rosen(x)
f2 = lambda x: 0.5*rosen(x) + 1.0*x[0] + .1*x[1] -1.0
#f = lambda x: x[0]**2*x[1]/20+.08
#f2 = lambda x: x[0]**2*x[1]/20. + (x[0]+x[1])/8.+.3
#f2 = lambda x: (x[0]+x[1]-5)**2/30 + (x[0]-x[1]-12)**2/120 - 3
#f = lambda x: x[0]**2*x[1]/20. - 1
#f2 = lambda x: x[0]**2*x[1]/20. + (x[0]+x[1])/5. - 2
# font = {'family' : 'normal',
# 'weight' : 'normal',
# 'size' : 14}
#
# rc('font', **font)
#f2 = lambda x: np.exp(x[0]/3)+np.exp(x[1]/5)-x[0]
lb = [-2,-2]
ub = [2,2]
dx = 0.1
x = y = np.arange(lb[0],ub[0],dx)
X, Y = np.meshgrid(x,y)
zs1 = np.array([f([x,y]) for x,y in zip(np.ravel(X),np.ravel(Y))])
Z1 = zs1.reshape(X.shape)
zs2 = np.array([f2([x,y]) for x,y in zip(np.ravel(X),np.ravel(Y))])
Z2 = zs2.reshape(X.shape)
#fig = plt.figure(1)
#ax1 = fig.add_subplot(111)
#ax1 = Axes3D(fig)
#ax1.plot_wireframe(X,Y,Z1)
# cs1 = ax1.contour(X,Y,Z1,levels=np.arange(0,20,1),colors='r')
# cs1 = ax1.contour(X,Y,Z2,levels=np.arange(0,20,1),colors='b')
# plt.clabel(cs1)
fig1 = plt.figure(1)
ax1 = Axes3D(fig1)
ax1.hold(True)
ax1.plot_wireframe(X,Y,Z1,color='r')
#ax1.contour(X,Y,Z1,color='r',levels=[0])
#ax1.plot_wireframe(X,Y,Z2,color='b')
#ax1.legend(['High fidelity function','Low fidelity function'])
ax1.set_xlabel('x1')
ax1.set_ylabel('x2')
ax1.set_zlabel('f (x1,x2)')
#ax1.axis([0,10,0,10])
ax1.set_zbound(-500,4000)
fig2 = plt.figure(2)
ax2 = Axes3D(fig2)
ax2.hold(True)
# #ax2.plot_surface(X,Y,Z1,color='r')
ax2.plot_wireframe(X,Y,Z2,color='b')
#ax1.contour(X,Y,Z2,color='b',levels=[0])
#ax2.legend(['High fidelity function','Low fidelity function'])
ax2.set_xlabel('x1')
ax2.set_ylabel('x2')
ax2.set_zlabel('f (x1,x2)')
# ax2.axis([0,10,0,10])
ax2.set_zbound(-500,4000)
plt.show()
def rosen_obj_func(w):
return rosen(w.T.toarray()[0])
def fitness(self, individual):
return rosen(individual)
def compute_r(self):
result = np.array((rosen(self.x.r) ))
return result
def my_rosen(x):
return rosen(x.T)
def my_rosen(x):
return rosen(x.T)
def myfunc1(x):
return optimize.rosen(x)
def rosenbrock(vector):
if all(-30 <= e <= 30 for e in vector):
return optimize.rosen(vector)
else:
return float("inf")
def f(x):
#return np.power(x[0], 2.) + np.power(x[1], 2.)
return optimize.rosen(x) # The Rosenbrock function
fhigh1d = lambda x: (6.0*x-2.0)**2.0*np.sin(12.*x-4.)
def fhigh(x):
A = 0.07
B = 0.9
C = 0.2
return fhigh1d(np.linalg.norm(A*x)) + B*np.sum(x) + C
def flow(x):
A = 0.035
B = 0.5
C = 0.1
return fhigh1d(np.linalg.norm(A*x)) + B*np.sum(x) + C
fhigh = lambda x: rosen(x)
flow = lambda x: 0.5*rosen(x) + 1.0*x[0] + .1*x[1] -1.0
g1high = lambda x: x[0]**2*x[1]/20. -.1
g1low = lambda x: x[0]**2*x[1]/20. + (x[0]+x[1])/5. -2
g2 = lambda x: (x[0]+x[1]-5)**2/30 + (x[0]-x[1]-12)**2/120 - 1
#fhigh = lambda x: 4.*x[0]**2. + x[1]**3. + x[0]*x[1]
#flow = lambda x: 4.*(x[0]+0.1)**2. + (x[1]-0.1)**3. + x[0]*x[1]+0.1
#
#g1high = lambda x: -(1./x[0]+ 1/x[1] -2.)
#g1low = lambda x: -(1./x[0] + 1./(x[1]+0.1)-2.001)
def problem_2d():
cnstr = ({'type':'ineq','fun':g1high},{'type':'ineq','fun':g2})
optlog.callback(x0)
try:
xfin = opt.minimize(f, jac=True, x0=x0, method='CG', callback=optlog.callback)
xfin = opt.minimize(f, jac=True, x0=xfin.x, method='CG', callback=optlog.callback)
xfin = opt.minimize(f, jac=True, x0=xfin.x, method='CG', callback=optlog.callback)
finx = xfin.x
optlog.finish(finx)
print(xfin)
except KeyboardInterrupt:
finx = optlog.hist.iloc[-1, :].x
print(finx)
hist = pd.read_pickle('./opthist.pkl')
plt.figure()
plt.plot(hist.i, hist.f, '-x')
plt.plot(hist.i, hist.gnorm, '-x')
plt.xscale('log')
plt.yscale('log')
plt.figure()
X, Y = np.meshgrid(np.linspace(-5, 5, 500),
np.linspace(-5, 5, 500))
xy = np.hstack((X.flatten()[:, None], Y.flatten()[:, None]))
feval = np.array([opt.rosen(v) for v in xy]).reshape(len(X), len(Y))
plt.contour(X, Y, np.log(feval))
hist_points = np.vstack(hist.x)
plt.plot(hist_points[:, 0], hist_points[:, 1], 'x-')
plt.show()
def f(self, x):
return rosen(x)
def test_rosen(var,*args):
import scipy.optimize as opt
return opt.rosen(var)
def f(x, g):
g[:] = rosen_der(x)
print "one call"
return rosen(x)
