def test_branin_2D_mixed_parallel(self):
n_parallel = 5
n_iter = 20
fun = Branin(ndim=2)
xlimits = fun.xlimits
criterion = "EI" #'EI' or 'SBO' or 'LCB'
qEI = "KB"
xtypes = [ORD, FLOAT]
sm = KRG(print_global=False)
mixint = MixedIntegerContext(xtypes, xlimits)
sampling = mixint.build_sampling_method(FullFactorial)
xdoe = sampling(10)
ego = EGO(
xdoe=xdoe,
n_iter=n_iter,
criterion=criterion,
xtypes=[ORD, FLOAT],
xlimits=xlimits,
n_parallel=n_parallel,
qEI=qEI,
evaluator=ParallelEvaluator(),
surrogate=sm,
random_state=42,
)
x_opt, y_opt, _, _, _ = ego.optimize(fun=fun)
# 3 optimal points possible: [-pi, 12.275], [pi, 2.275], [9.42478, 2.475]
self.assertTrue(
np.allclose([[-3, 12.275]], x_opt, rtol=0.2)
or np.allclose([[3, 2.275]], x_opt, rtol=0.2)
or np.allclose([[9, 2.475]], x_opt, rtol=0.2))
self.assertAlmostEqual(0.494, float(y_opt), delta=1)
def _setup_optimizer(self, fun):
"""
Instanciate internal surrogate used for optimization
and setup function evaluator wrt options
Parameters
----------
fun: function to optimize: ndarray[n, nx] or ndarray[n] -> ndarray[n, 1]
Returns
-------
ndarray: initial coord-x doe
ndarray: initial coord-y doe = fun(xdoe)
"""
# Set the model
self.gpr = self.options["surrogate"]
self.xlimits = self.options["xlimits"]
# Handle mixed integer optimization
xtypes = self.options["xtypes"]
if xtypes:
self.mixint = MixedIntegerContext(xtypes,
self.xlimits,
work_in_folded_space=False)
self.gpr = self.mixint.build_surrogate_model(self.gpr)
self._sampling = self.mixint.build_sampling_method(
LHS,
criterion="ese",
random_state=self.options["random_state"])
else:
self.mixint = None
self._sampling = LHS(
xlimits=self.xlimits,
criterion="ese",
random_state=self.options["random_state"],
)
# Build DOE
self._evaluator = self.options["evaluator"]
xdoe = self.options["xdoe"]
if xdoe is None:
self.log("Build initial DOE with LHS")
n_doe = self.options["n_doe"]
x_doe = self._sampling(n_doe)
else:
self.log("Initial DOE given")
x_doe = np.atleast_2d(xdoe)
if self.mixint:
x_doe = self.mixint.unfold_with_enum_mask(x_doe)
ydoe = self.options["ydoe"]
if ydoe is None:
y_doe = self._evaluator.run(fun, x_doe)
else: # to save time if y_doe is already given to EGO
y_doe = ydoe
return x_doe, y_doe
def test_krg_mixed_3D_bad_regr(self):
xtypes = [FLOAT, (ENUM, 3), ORD]
xlimits = [[-10, 10], ["blue", "red", "green"], [-10, 10]]
mixint = MixedIntegerContext(xtypes, xlimits)
with self.assertRaises(ValueError):
sm = mixint.build_surrogate_model(
KRG(print_prediction=False, poly="linear")
)
def test_ego_mixed_integer_full_gaussian(self):
n_iter = 15
xtypes = [FLOAT, (ENUM, 3), (ENUM, 2), ORD]
xlimits = np.array([[-5, 5], ["blue", "red", "green"],
["large", "small"], [0, 2]])
n_doe = 2
sampling = MixedIntegerSamplingMethod(
xtypes,
xlimits,
LHS,
criterion="ese",
random_state=42,
output_in_folded_space=True,
)
xdoe = sampling(n_doe)
criterion = "EI" #'EI' or 'SBO' or 'LCB'
sm = KRG(print_global=False)
mixint = MixedIntegerContext(xtypes, xlimits)
ego = EGO(
n_iter=n_iter,
criterion=criterion,
xdoe=xdoe,
xtypes=xtypes,
xlimits=xlimits,
surrogate=sm,
enable_tunneling=False,
random_state=42,
categorical_kernel=FULL_GAUSSIAN,
)
_, y_opt, _, _, _ = ego.optimize(
fun=TestEGO.function_test_mixed_integer)
def test_ego_mixed_integer(self):
n_iter = 15
xtypes = [FLOAT, (ENUM, 3), (ENUM, 2), ORD]
xlimits = np.array([[-5, 5], ["blue", "red", "green"],
["large", "small"], ["0", "2", "3"]])
n_doe = 2
sampling = MixedIntegerSamplingMethod(xtypes,
xlimits,
LHS,
criterion="ese",
random_state=42)
xdoe = sampling(n_doe)
criterion = "EI" #'EI' or 'SBO' or 'LCB'
sm = KRG(print_global=False)
mixint = MixedIntegerContext(xtypes, xlimits)
ego = EGO(
n_iter=n_iter,
criterion=criterion,
xdoe=xdoe,
xtypes=xtypes,
xlimits=xlimits,
surrogate=sm,
enable_tunneling=False,
random_state=42,
)
_, y_opt, _, _, _ = ego.optimize(
fun=TestEGO.function_test_mixed_integer)
self.assertAlmostEqual(-15, float(y_opt), delta=5)
def test_branin_2D_mixed(self):
n_iter = 20
fun = Branin(ndim=2)
xtypes = [INT, FLOAT]
xlimits = fun.xlimits
criterion = "EI" #'EI' or 'SBO' or 'UCB'
sm = KRG(print_global=False)
mixint = MixedIntegerContext(xtypes, xlimits)
sampling = MixedIntegerSamplingMethod(xtypes, xlimits, FullFactorial)
xdoe = sampling(10)
ego = EGO(
xdoe=xdoe,
n_iter=n_iter,
criterion=criterion,
xtypes=xtypes,
xlimits=xlimits,
surrogate=sm,
random_state=42,
)
x_opt, y_opt, _, _, _ = ego.optimize(fun=fun)
# 3 optimal points possible: [-pi, 12.275], [pi, 2.275], [9.42478, 2.475]
self.assertTrue(
np.allclose([[-3, 12.275]], x_opt, rtol=0.2)
or np.allclose([[3, 2.275]], x_opt, rtol=0.2)
or np.allclose([[9, 2.475]], x_opt, rtol=0.2))
self.assertAlmostEqual(0.494, float(y_opt), delta=1)
def test_qp_mixed_2D_INT(self):
xtypes = [FLOAT, INT]
xlimits = [[-10, 10], [-10, 10]]
mixint = MixedIntegerContext(xtypes, xlimits)
sm = mixint.build_surrogate_model(QP(print_prediction=False))
sampling = mixint.build_sampling_method(LHS, criterion="m")
fun = Sphere(ndim=2)
xt = sampling(10)
yt = fun(xt)
sm.set_training_values(xt, yt)
sm.train()
eq_check = True
for i in range(xt.shape[0]):
if abs(float(xt[i, :][1]) - int(float(xt[i, :][1]))) > 10e-8:
eq_check = False
self.assertTrue(eq_check)
def test_krg_mixed_3D(self):
xtypes = [FLOAT, (ENUM, 3), INT]
xlimits = [[-10, 10], ["blue", "red", "green"], [-10, 10]]
mixint = MixedIntegerContext(xtypes, xlimits)
sm = mixint.build_surrogate_model(KRG(print_prediction=False))
sampling = mixint.build_sampling_method(LHS, criterion="m")
fun = Sphere(ndim=3)
xt = sampling(20)
yt = fun(xt)
sm.set_training_values(xt, yt)
sm.train()
eq_check = True
for i in range(xt.shape[0]):
if abs(float(xt[i, :][2]) - int(float(xt[i, :][2]))) > 10e-8:
eq_check = False
if not (xt[i, :][1] == 0 or xt[i, :][1] == 1 or xt[i, :][1] == 2):
eq_check = False
self.assertTrue(eq_check)
def run_mixed_integer_context_example(self):
import numpy as np
import matplotlib.pyplot as plt
from matplotlib import colors
from mpl_toolkits.mplot3d import Axes3D
from smt.surrogate_models import KRG
from smt.sampling_methods import LHS, Random
from smt.applications.mixed_integer import MixedIntegerContext, FLOAT, INT, ENUM
xtypes = [INT, FLOAT, (ENUM, 4)]
xlimits = [[0, 5], [0.0, 4.0], ["blue", "red", "green", "yellow"]]
def ftest(x):
return (x[:, 0] * x[:, 0] + x[:, 1] * x[:, 1]) * (x[:, 2] + 1)
# context to create consistent DOEs and surrogate
mixint = MixedIntegerContext(xtypes, xlimits)
# DOE for training
lhs = mixint.build_sampling_method(LHS, criterion="ese")
num = mixint.get_unfolded_dimension() * 5
print("DOE point nb = {}".format(num))
xt = lhs(num)
yt = ftest(xt)
# Surrogate
sm = mixint.build_surrogate_model(KRG())
print(xt)
sm.set_training_values(xt, yt)
sm.train()
# DOE for validation
rand = mixint.build_sampling_method(Random)
xv = rand(50)
yv = ftest(xv)
yp = sm.predict_values(xv)
plt.plot(yv, yv)
plt.plot(yv, yp, "o")
plt.xlabel("actual")
plt.ylabel("prediction")
plt.show()
def test_ego_mixed_integer(self):
n_iter = 15
xlimits = np.array([[-5, 5], ["1", "2", "3"], ["4", "5"]],
dtype="object")
xdoe = np.array([[5, 0, 0], [4, 0, 0]])
criterion = "EI" #'EI' or 'SBO' or 'UCB'
xtypes = [INT, (ENUM, 3), (ENUM, 2)]
sm = KRG(print_global=False)
mixint = MixedIntegerContext(xtypes, xlimits)
ego = EGO(
n_iter=n_iter,
criterion=criterion,
xdoe=xdoe,
xtypes=xtypes,
xlimits=xlimits,
surrogate=sm,
enable_tunneling=False,
random_state=42,
)
_, y_opt, _, _, _ = ego.optimize(
fun=TestEGO.function_test_mixed_integer)
self.assertAlmostEqual(-15, float(y_opt), delta=5)
class EGO(SurrogateBasedApplication):
def _initialize(self):
super(EGO, self)._initialize()
declare = self.options.declare
declare("fun", None, types=FunctionType, desc="Function to minimize")
declare(
"criterion",
"EI",
types=str,
values=["EI", "SBO", "UCB"],
desc=
"criterion for next evaluation point determination: Expected Improvement, \
Surrogate-Based Optimization or Upper Confidence Bound",
)
declare("n_iter", None, types=int, desc="Number of optimizer steps")
declare(
"n_max_optim",
20,
types=int,
desc="Maximum number of internal optimizations",
)
declare("n_start",
20,
types=int,
desc="Number of optimization start points")
declare(
"n_parallel",
1,
types=int,
desc="Number of parallel samples to compute using qEI criterion",
)
declare(
"qEI",
"KBLB",
types=str,
values=["KB", "KBLB", "KBUB", "KBRand", "CLmin"],
desc="Approximated q-EI maximization strategy",
)
declare(
"evaluator",
default=Evaluator(),
types=Evaluator,
desc=
"Object used to run function fun to optimize at x points (nsamples, nxdim)",
)
declare(
"n_doe",
None,
types=int,
desc=
"Number of points of the initial LHS doe, only used if xdoe is not given",
)
declare("xdoe", None, types=np.ndarray, desc="Initial doe inputs")
declare("ydoe", None, types=np.ndarray, desc="Initial doe outputs")
declare("xlimits",
None,
types=np.ndarray,
desc="Bounds of function fun inputs")
declare("verbose",
False,
types=bool,
desc="Print computation information")
declare(
"enable_tunneling",
False,
types=bool,
desc=
"Enable the penalization of points that have been already evaluated in EI criterion",
)
declare(
"surrogate",
KRG(print_global=False),
types=(KRG, KPLS, KPLSK, MGP),
desc="SMT kriging-based surrogate model used internaly",
)
declare(
"xtypes",
None,
types=list,
desc=
"x type specifications: either FLOAT for continuous, INT for integer "
"or (ENUM n) for categorical doimension with n levels",
)
self.options.declare(
"random_state",
types=(type(None), int, np.random.RandomState),
desc=
"Numpy RandomState object or seed number which controls random draws",
)
def optimize(self, fun):
"""
Optimizes fun
Parameters
----------
fun: function to optimize: ndarray[n, nx] or ndarray[n] -> ndarray[n, 1]
Returns
-------
[nx, 1]: x optimum
[1, 1]: y optimum
int: index of optimum in data arrays
[ndoe + n_iter, nx]: coord-x data
[ndoe + n_iter, 1]: coord-y data
[ndoe, nx]: coord-x initial doe
[ndoe, 1]: coord-y initial doe
"""
x_data, y_data = self._setup_optimizer(fun)
n_iter = self.options["n_iter"]
n_parallel = self.options["n_parallel"]
for k in range(n_iter):
# Virtual enrichement loop
for p in range(n_parallel):
# find next best x-coord point to evaluate
x_et_k, success = self._find_best_point(
x_data, y_data, self.options["enable_tunneling"])
if not success:
self.log(
"Internal optimization failed at EGO iter = {}.{}".
format(k, p))
break
elif success:
self.log(
"Internal optimization succeeded at EGO iter = {}.{}".
format(k, p))
# Set temporaly the y-coord point based on the kriging prediction
y_et_k = self._get_virtual_point(np.atleast_2d(x_et_k), y_data)
# Update y_data with predicted value
y_data = np.atleast_2d(np.append(y_data, y_et_k)).T
x_data = np.atleast_2d(np.append(x_data, x_et_k, axis=0))
# Compute the real values of y_data
x_to_compute = np.atleast_2d(x_data[-n_parallel:])
if self.mixint:
x_to_compute = self.mixint.fold_with_enum_index(x_to_compute)
y = self._evaluator.run(fun, x_to_compute)
y_data[-n_parallel:] = y
# Find the optimal point
ind_best = np.argmin(y_data)
x_opt = x_data[ind_best]
y_opt = y_data[ind_best]
if self.mixint:
x_opt = self.mixint.fold_with_enum_index(x_opt)[0]
return x_opt, y_opt, ind_best, x_data, y_data
def log(self, msg):
if self.options["verbose"]:
print(msg)
def EI(self, points, y_data, enable_tunneling=False, x_data=None):
""" Expected improvement """
f_min = np.min(y_data)
pred = self.gpr.predict_values(points)
sig = np.sqrt(self.gpr.predict_variances(points))
args0 = (f_min - pred) / sig
args1 = (f_min - pred) * norm.cdf(args0)
args2 = sig * norm.pdf(args0)
if sig.size == 1 and sig == 0.0: # can be use only if one point is computed
return 0.0
ei = args1 + args2
# penalize the points already evaluated with tunneling
if enable_tunneling:
for i in range(len(points)):
p = np.atleast_2d(points[i])
EIp = self.EI(p, y_data, enable_tunneling=False)
for x in x_data:
x = np.atleast_2d(x)
# if np.abs(p-x)<1:
# ei[i]=ei[i]*np.reciprocal(1+100*np.exp(-np.reciprocal(1-np.square(p-x))))
pena = (EIp - self.EI(x, y_data, enable_tunneling=False)
) / np.power(np.linalg.norm(p - x), 4)
if pena > 0:
ei[i] = ei[i] - pena
ei[i] = max(ei[i], 0)
return ei
def SBO(self, point):
""" Surrogate based optimization: min the surrogate model by suing the mean mu """
res = self.gpr.predict_values(point)
return res
def UCB(self, point):
""" Upper confidence bound optimization: minimize by using mu - 3*sigma """
pred = self.gpr.predict_values(point)
var = self.gpr.predict_variances(point)
res = pred - 3.0 * np.sqrt(var)
return res
def _setup_optimizer(self, fun):
"""
Instanciate internal surrogate used for optimization
and setup function evaluator wrt options
Parameters
----------
fun: function to optimize: ndarray[n, nx] or ndarray[n] -> ndarray[n, 1]
Returns
-------
ndarray: initial coord-x doe
ndarray: initial coord-y doe = fun(xdoe)
"""
# Set the model
self.gpr = self.options["surrogate"]
self.xlimits = self.options["xlimits"]
# Handle mixed integer optimization
xtypes = self.options["xtypes"]
if xtypes:
self.mixint = MixedIntegerContext(xtypes,
self.xlimits,
work_in_folded_space=False)
self.gpr = self.mixint.build_surrogate_model(self.gpr)
self._sampling = self.mixint.build_sampling_method(
LHS,
criterion="ese",
random_state=self.options["random_state"])
else:
self.mixint = None
self._sampling = LHS(
xlimits=self.xlimits,
criterion="ese",
random_state=self.options["random_state"],
)
# Build DOE
self._evaluator = self.options["evaluator"]
xdoe = self.options["xdoe"]
if xdoe is None:
self.log("Build initial DOE with LHS")
n_doe = self.options["n_doe"]
x_doe = self._sampling(n_doe)
else:
self.log("Initial DOE given")
x_doe = np.atleast_2d(xdoe)
if self.mixint:
x_doe = self.mixint.unfold_with_enum_mask(x_doe)
ydoe = self.options["ydoe"]
if ydoe is None:
y_doe = self._evaluator.run(fun, x_doe)
else: # to save time if y_doe is already given to EGO
y_doe = ydoe
return x_doe, y_doe
def _find_best_point(self,
x_data=None,
y_data=None,
enable_tunneling=False):
"""
Function that analyse a set of x_data and y_data and give back the
more interesting point to evaluates according to the selected criterion
Parameters
----------
x_data: ndarray(n_points, nx)
y_data: ndarray(n_points, 1)
Returns
-------
ndarray(nx, 1): the next best point to evaluate
boolean: success flag
"""
self.gpr.set_training_values(x_data, y_data)
self.gpr.train()
criterion = self.options["criterion"]
n_start = self.options["n_start"]
n_max_optim = self.options["n_max_optim"]
if self.mixint:
bounds = self.mixint.get_unfolded_xlimits()
else:
bounds = self.xlimits
if criterion == "EI":
self.obj_k = lambda x: -self.EI(np.atleast_2d(x), y_data,
enable_tunneling, x_data)
elif criterion == "SBO":
self.obj_k = lambda x: self.SBO(np.atleast_2d(x))
elif criterion == "UCB":
self.obj_k = lambda x: self.UCB(np.atleast_2d(x))
success = False
n_optim = 1 # in order to have some success optimizations with SLSQP
while not success and n_optim <= n_max_optim:
opt_all = []
x_start = self._sampling(n_start)
for ii in range(n_start):
try:
opt_all.append(
minimize(
lambda x: float(self.obj_k(x)),
x_start[ii, :],
method="SLSQP",
bounds=bounds,
options={"maxiter": 200},
))
except ValueError: # in case "x0 violates bound constraints" error
print("warning: `x0` violates bound constraints")
print("x0={}".format(x_start[ii, :]))
print("bounds={}".format(bounds))
opt_all.append({"success": False})
opt_all = np.asarray(opt_all)
opt_success = opt_all[[opt_i["success"] for opt_i in opt_all]]
obj_success = np.array([opt_i["fun"] for opt_i in opt_success])
success = obj_success.size != 0
if not success:
self.log("New start point for the internal optimization")
n_optim += 1
if n_optim >= n_max_optim:
# self.log("Internal optimization failed at EGO iter = {}".format(k))
return np.atleast_2d(0), False
ind_min = np.argmin(obj_success)
opt = opt_success[ind_min]
x_et_k = np.atleast_2d(opt["x"])
return x_et_k, True
def _get_virtual_point(self, x, y_data):
"""
Depending on the qEI attribute return a predicted value at given point x
Parameters
----------
x: ndarray(1, 1) the x-coord point where to forecast the y-coord virtual point
y_data: current y evaluation list only used when qEI is CLmin
Returns
-------
ndarray(1, 1): the so-called virtual y-coord point
"""
qEI = self.options["qEI"]
if qEI == "CLmin":
return np.min(y_data)
if qEI == "KB":
return self.gpr.predict_values(x)
if qEI == "KBUB":
conf = 3.0
if qEI == "KBLB":
conf = -3.0
if qEI == "KBRand":
conf = np.random.randn()
pred = self.gpr.predict_values(x)
var = self.gpr.predict_variances(x)
return pred + conf * np.sqrt(var)
def run_ego_mixed_integer_example():
import numpy as np
from smt.applications import EGO
from smt.applications.mixed_integer import (
MixedIntegerContext,
FLOAT,
ENUM,
ORD,
)
import matplotlib.pyplot as plt
from smt.surrogate_models import KRG
from smt.sampling_methods import LHS
# Regarding the interface, the function to be optimized should handle
# categorical values as index values in the enumeration type specification.
# For instance, here "blue" will be passed to the function as the index value 2.
# This allows to keep the numpy ndarray X handling numerical values.
def function_test_mixed_integer(X):
# float
x1 = X[:, 0]
# enum 1
c1 = X[:, 1]
x2 = c1 == 0
x3 = c1 == 1
x4 = c1 == 2
# enum 2
c2 = X[:, 2]
x5 = c2 == 0
x6 = c2 == 1
# int
i = X[:, 3]
y = ((x2 + 2 * x3 + 3 * x4) * x5 * x1 +
(x2 + 2 * x3 + 3 * x4) * x6 * 0.95 * x1 + i)
return y
n_iter = 15
xtypes = [FLOAT, (ENUM, 3), (ENUM, 2), ORD]
xlimits = np.array([[-5, 5], ["red", "green", "blue"],
["square", "circle"], [0, 2]])
criterion = "EI" #'EI' or 'SBO' or 'LCB'
qEI = "KB"
sm = KRG(print_global=False)
mixint = MixedIntegerContext(xtypes, xlimits)
n_doe = 3
sampling = mixint.build_sampling_method(LHS,
criterion="ese",
random_state=42)
xdoe = sampling(n_doe)
ydoe = function_test_mixed_integer(xdoe)
ego = EGO(
n_iter=n_iter,
criterion=criterion,
xdoe=xdoe,
ydoe=ydoe,
xtypes=xtypes,
xlimits=xlimits,
surrogate=sm,
qEI=qEI,
random_state=42,
)
x_opt, y_opt, _, _, y_data = ego.optimize(
fun=function_test_mixed_integer)
print("Minimum in x={} with f(x)={:.1f}".format(x_opt, float(y_opt)))
print("Minimum in typed x={}".format(
ego.mixint.cast_to_mixed_integer(x_opt)))
min_ref = -15
mini = np.zeros(n_iter)
for k in range(n_iter):
mini[k] = np.log(np.abs(np.min(y_data[0:k + n_doe - 1]) - min_ref))
x_plot = np.linspace(1, n_iter + 0.5, n_iter)
u = max(np.floor(max(mini)) + 1, -100)
l = max(np.floor(min(mini)) - 0.2, -10)
fig = plt.figure()
axes = fig.add_axes([0.1, 0.1, 0.8, 0.8])
axes.plot(x_plot, mini, color="r")
axes.set_ylim([l, u])
plt.title("minimum convergence plot", loc="center")
plt.xlabel("number of iterations")
plt.ylabel("log of the difference w.r.t the best")
plt.show()
